I configured a LAG into a stack composed by 2 x SG550X-48 (Units 1/2) and 2 x SG550X-24 (Units 3/4).
The LAG1 is configured with LACP on GE3/4 and GE4/4 ports.
The problem is that the GE4/4 port is recognised as a standby member:
even if the link is UP:
and the operating system (Debian Linux Stretch) recognises it as up and running with both interfaces:
For this reason I have about 50% of packet loss to and from this server and this is a serious problem.
If I try to disable the LACP I don't have the problem:
The strange thing is that I've already used the same configuration in another environment and that works good.
You're trying to make an mlag. Even in identical widely used hardware, support for this feature is spotty even today. Furthermore, mlag isn't an agreed upon standard, and varies functionally across different devices and software versions. If you're joining two identical switches via mlag, you'll probably end up fine.
Check your firmware versions between this problem child and your working switch stack. If there are any discrepancies, see if bringing them to the same version helps.
Also look into approaching this differently if this doesn't work out or you plan on using SDN now or in the future (mlag support via SDN is just about non-existent). You might try a combination of ECMP and BGP, which is a well-tested pairing that would potentially address your desires at layer 3 rather than wholly at layer 2.