FX gets called a 24 hour market. That's roughly true from Sunday evening to Friday evening New York time, and much less true of the connections underneath it. Around the daily rollover at 5pm New York, and again at the weekend open, a lot of LP sessions reset, quotes thin out, and the bridge finds out whether its recovery logic is any good.
Sequence numbers come first. Many LPs reset their FIX sequence numbers at a fixed time each day, and some want ResetSeqNumFlag (141=Y) on the logon. If the bridge kept its numbers from before the reset and the LP didn't, you get the familiar 'MsgSeqNum too low' logout and a reconnect loop.
Get the session schedule and reset policy for each LP and keep them in config per LP, because they don't all behave the same. The times are often defined in New York time, so the UTC hour moves when the US changes its clocks, and for a few weeks a year the US and Europe are on different offsets. A hardcoded UTC time will get caught out by that.
Quotes that look alive
Over the weekend the last prices from Friday are still sitting in the bridge unless something clears them. If the platform keeps showing them and accepts orders against them, you have a stale quote problem that only happens once a week, which is why it gets missed in testing. When prices do come back on Sunday they can be a long way from Friday's close, with wide spreads. A bridge that opens trading as soon as it has one quote from one LP will publish something strange.
A safer pattern is to keep trading closed per symbol until the bridge has seen several fresh quotes from more than one LP, those quotes agree with each other within some tolerance, and the spread has come under a limit. The right numbers for 'several' and 'tolerance' depend on the pair and the book, so they should be settings you tune, not values copied from somewhere else.
Rollover itself
Around 5pm New York liquidity thins out for a while and spreads widen. If the bridge has a max spread filter it may trip and pause trading at a time when clients in some regions are very active. Whether that's what you want is a business call, but it should be a call somebody made. It's also when MT applies swaps, and the MT server runs on its own clock. What the bridge treats as rollover and what the server treats as rollover need to be the same moment, or positions near the cutoff get a swap they shouldn't, or miss one they should.
The most useful test is probably the dullest one: run the bridge through a whole weekend against LP test sessions, if they offer them, and read the logs on Monday. It's slow work and it finds problems unit tests don't.