I noticed, that failed host (lost datastore) is allways one on witch vcenter appliance is running.

Workaround was, I reinstall vcenter on third host that is not part of the cluster, only member of datacenter and has only local storage (not msa).

That is obviously unrelated from the storage issue and just a coincidence at best. HA works independent from vCenter, even if vCenter is down. You only need an available vCenter server to make changes to the HA configuration.

The worst thing is, ha agent not restarts vms on another host untill I physically turn that host off.

Well, that's exactly what HA is supposed to do: Restart VMs in case of a host failure. HA does not protect against and was never supposed to handle storage-side failures.

You need to dig down to the cause of the storage issues instead of blaming the wrong effect.