I've noticed that during CallManager failover, when we reboot the publisher, one of our MGCP gateways often reset all of its PRI connections. It's very odd for two reasons:
1) The publisher (10.1.2.10) isn't the active call agent. The subscriber is.
2) A similar gateway with the same IOS version and configuration, but less PRIs does not have this problem.
I've tried a few different IOS versions, and that doesn't seem to help. Also it has happened with both CCM 4.1(3)sr1c and 4.1(3)sr4d.