What controls SenderDisconnects on an Azure Relay

Question

We have an Azure API that connects to an on-prem Oracle database, and by accessing the metrics of the relay we can see how many Active connections we have open, and how many SenderDisconnects have been issued.

Performance of these api calls has been an issue for some time, and whereas we will typically have sub-second response times, regularly for a period of maybe 5 minutes this will slow right down, and standard api calls will take over 2 minutes (which is our timeout threshold). Have noticed that these times correspond to sender disconnects being issued, but can't find too much about them in MS documentation. So my questions are

1. What is issuing the SenderDisconnect instruction? Is it the api itself that is unhooking itself from the relay
2. Why is it doing this? I originally thought it was when it hit a limit of connections, but this doesn't seem to be the case - is it caused by a period of inactivity?
3. Is there anyway to control this, so we could (e.g.) flush the connections before the working day and force them to stay open?