I think we might be missing the forest for the trees here, guys.
MT say that the reduction here shouldn't make a difference. In this post
by strods from page 7, strods says that neither of the routerOS RADIUS client processes will wait more than 3 seconds even if the timeout setting is greater than 3 seconds
- important part in bold for emphasis;
RADIUS timeout value was reduced due to the fact that there is no point of higher value than 3s. Neither of RouterOS RADIUS services would wait more than 3s for a reply from RADIUS server. If you had value higher than 3 seconds, then either configuration will work with timeout set to 3s or it was not working properly even with higher value.
Anyway, RADIUS timeout (even set to 3s) is too high and requires debugging and network improvements since router should not wait so long for a reply from RADIUS server.
Ignoring the part about RADIUS taking more than 3 seconds being unacceptable for the time being, as described by strods, this change should not have resulted in a regression of functionality, as the RADIUS client inside routerOS was not honoring timeouts above 3 seconds to begin with. Based on what mducharme says above, this does not appear to be the case.
He reports that he's seeing different behavior with the timeout hard-set to 3sec, when compared to the 6sec available before. This is a regression in functionality and directly contrary to the information given by MT (strods) - in short, this is a bug.
Either MT themselves are incorrect about the behaviour of the client when given timeouts above 3s (and mducharme's 6s timeout on older FW versions was
making a difference), or something else has changed that otherwise breaks his admittedly unusual but completely valid configuration.
As mducharme has demonstrated, while unusual, it's not unreasonable for RADIUS to take some seconds to reply, and alternate vendors appear to have max timeouts in the 30-90 second range at the very least. We can argue about arbitrary configuration constraints until we're blue in the face, but it's not relevant to the issue at hand here.
MT: What's your explanation for why mducharme's configuration here works on 6.40 but not on 6.41?
mducharme: Have you opened a support ticket for this regression & provided supout.rifs etc? This isn't an official support channel, and it sounds like you should be talking to support if you aren't already.