So we figured it out what was the problem - we had 2 hosts, one of them was uninitialized virsh host (cores:0, memory:0):

root@maas:/home/ubuntu# maas admin machines allocate
list index out of range
root@maas:/home/ubuntu# maas admin vm-hosts read
Success.
Machine-readable output follows:
[
    {
        "default_macvlan_mode": "",
        "capabilities": [
            "composable",
            "dynamic_local_storage",
            "over_commit",
            "storage_pools"
        ],
        "host": {
            "system_id": "kwxmgm",
            "__incomplete__": true
        },
        "total": {
            "cores": 4,
            "memory": 8192,
            "local_storage": 22884122624
        },
        "storage_pools": [
            {
                "id": "default",
                "name": "default",
                "type": "zfs",
                "path": "/var/snap/lxd/common/lxd/disks/default.img",
                "total": 22884122624,
                "used": 0,
                "available": 22884122624,
                "default": true
            }
        ],
        "type": "lxd",
        "id": 1,
        "pool": {
            "name": "default",
            "description": "Default pool",
            "id": 0,
            "resource_uri": "/MAAS/api/2.0/resourcepool/0/"
        },
        "architectures": [
            "amd64/generic"
        ],
        "zone": {
            "name": "default",
            "description": "",
            "id": 1,
            "resource_uri": "/MAAS/api/2.0/zones/default/"
        },
        "memory_over_commit_ratio": 1.0,
        "available": {
            "cores": 4,
            "memory": 8192,
            "local_storage": 22884122624
        },
        "name": "crisp-skunk",
        "cpu_over_commit_ratio": 1.0,
        "used": {
            "cores": 0,
            "memory": 0,
            "local_storage": 0
        },
        "version": "5.6",
        "tags": [
            "pod-console-logging",
            "virtual"
        ],
        "resource_uri": "/MAAS/api/2.0/vm-hosts/1/"
    },
    {
        "default_macvlan_mode": null,
        "capabilities": [],
        "host": {
            "system_id": null,
            "__incomplete__": true
        },
        "total": {
            "cores": 0,
            "memory": 0,
            "local_storage": 0
        },
        "storage_pools": [],
        "type": "virsh",
        "id": 50,
        "pool": {
            "name": "default",
            "description": "Default pool",
            "id": 0,
            "resource_uri": "/MAAS/api/2.0/resourcepool/0/"
        },
        "architectures": [],
        "zone": {
            "name": "default",
            "description": "",
            "id": 1,
            "resource_uri": "/MAAS/api/2.0/zones/default/"
        },
        "memory_over_commit_ratio": 1.0,
        "available": {
            "cores": 0,
            "memory": 0,
            "local_storage": 0
        },
        "name": "sadsa",
        "cpu_over_commit_ratio": 1.0,
        "used": {
            "cores": 0,
            "memory": 0,
            "local_storage": 0
        },
        "version": "",
        "tags": [
            "pod-console-logging"
        ],
        "resource_uri": "/MAAS/api/2.0/vm-hosts/50/"
    }
]

After deleting virsh host, we were able to allocate new ephemeral machine on lxd host (version 5.6)