Odd STP related problem in V1910 with a stacked switch

[n3mmr]

I have a very straightforward network to deal with, a central switch made up of two stacked Edgecore ecs4510-28F switches feeding 11 V1910-48G switches.
The feeds are dynamic, LACP controlled, LAGs of 4*1G fibre SFPs
At the V1910 end the LAGs are made up of ports 49-52.
At the Edgecore end the LAGs are made up of two ports each from each stack half, e g port1+2 in switch1 + port1+2 in switch2.
The topology is a simple and straightforward star, the 4*1G LAGs connect each V1910 directly to the 4510 stack, and there are no other links between switches. The V1910 rj45 ports connect to individual flats, and the risk of a loop to appear between two flats seems minute, even with Apple style ethernet bridges and wifi leakage.

Now to the problem:
If I enable STP on the MSTP global configuration page in more than one V1910, all but one V1910 will eventually be cut off from the 4510stack.
All I do is enable globally, and all else is default.
What could possibly cause this?
There are no superfluous links, and the topology will never change, but I'm still uneasy about turning STP off, which would, I suppose, solve the problem.
What would be your first thought in a situation like mine?