COM

Client Application        <- Ole DB consumer application (MFC, .NET, Excel...)<br />
-----------------------<br />
OLE DB Provider           <- On the client. Dumb connection to my service via the Network layer<br />
-----------------------<br />
<br />
Network layer <- I really don't know how to pass through this<br />
<br />
-----------------------<br />
OLE DB Consumer/Provider service                  <- This should be on a network server. Here I'm implementing row level security. This service acts as an OLE DB consumer for the DB and exposes data to the client.<br />
-----------------------<br />
Physical DB               <- Any OLE DB-compliant database (mainly SQL Server or Oracle)<br />
-----------------------

_ConnectionPtr pConn;
// Some method in the app:
try
{
    pConn->Execute(...); // doing this while MySQL went kaboom leads to a _com_error being thrown
}
catch (_com_error &e)
{
    switch (e.Error())
    {
        case DB_E_ABORTLIMITREACHED:
            SetEvent(hReconnectDBEvent);
            break;
    }
...
}

// Connection thread:
while (keepChecking)
{
    waitRes = WaitForMultipleObjects(..., 30000); // App exit, DB reconnect, half the db connection timeout
    switch (waitRes)
    {
        case WAIT_OBJECT_the_reconnect_event:
            if (pConn->State == adStateOpen)
            {
                pConn->Close();
            }
            pConn->Open(...); // returns 0 when MySQL is back up
            
            break;
    }
}

Once connection is reopened ok attempts to Execute something still throw that exception for about 5 more minutes. After that everything returns back to normal.