We've had complaints from a couple users at Gates that they went to use Synapse only to find it was "down". The most recent one was from 10PM Saturday night, during our planned stack migration. So the issue is not a Synapse reliability/stability problem but rather one of user expectations: They did not expect to find Synapse unavailable when they went to do something, so they felt frustrated when they couldn't use it.
The challenge is to balance the need for flexibility by the operations team with the need for predictability by users. Today we give up the latter in favor of the former. Could we, for example, publish our 'best guess' as to the next scheduled down time, adjusting it as things change? If Synapse said, "Next scheduled down time, July 13, appx. 10:00PM PST", that might help our users know when not to plan to use the system. If things change we could update the date/time to reflect our best guess.