hosting/lib/docker_deploy.py

from lib import config as config_module
from lib import logging as logging_lib
from lib import docker_cli_wrapper
from lib import docker_interface
from lib import get_specs
import docker
from docker.types import EndpointConfig, NetworkingConfig
import os

client = docker_interface.client
config = config_module.config
log = logging_lib.log

def deploy(validated_containers, allowed_running_containers=[]):
    local_images = docker_interface.get_local_images()
    all_containers = docker_interface.get_containers(all=True)

    is_hive = "hive" in get_specs.get_kernel()

    # Deploy wireguard first

    wireguard_containers = []
    rest_containers = []
    for container in validated_containers:
        if not "name" in container or not "image" in container:
            pass
        elif "wireguard" in container:
            wireguard_containers.append(container)
        else:
            rest_containers.append(container)

    validated_containers = wireguard_containers+rest_containers

    # Deploy wireguard first

    needed_running_names = []
    paused_names = []

    created_container_names = []
    for container in all_containers:
        if type(container.name)==str:
            created_container_names.append(container.name)

    for validated_container in validated_containers:
        try:

            image_ready = False
            docker_gpus = None
            for local_image in local_images:
                if local_image.replace(':latest','')==validated_container["image"].replace(':latest',''):
                    image_ready = True
                    break

            if "paused" in validated_container:
                paused_names.append(validated_container["name"])
            else:
                needed_running_names.append(validated_container["name"])

            container_options = {
                'image': validated_container["image"],
                'name': validated_container["name"],
                'detach': True,
                'tty': True,
                'network_mode': 'clore-br0',
                'cap_add': [],
                'volumes': {},
                'ports': {},
                'device_requests': [],
                'environment': validated_container["env"] if "env" in validated_container else {},
                'log_config': docker.types.LogConfig(
                    type='json-file',
                    config={
                        'max-size': '5m',
                        'max-file': '1'
                    }
                )
            }

            if "hostname" in validated_container:
                container_options["hostname"]=validated_container["hostname"]
            elif "clore-order-" in validated_container["name"]:
                try:
                    container_options["hostname"] = f"O-{int(validated_container["name"][12:])}"
                except Exception as eon:
                    pass
            if "network" in validated_container:
                container_options["network_mode"]=validated_container["network"]
            if "ip" in validated_container and config.creation_engine=="sdk":
                del container_options["network_mode"]

            if "gpus" in validated_container and type(validated_container["gpus"])==bool:
                container_options["runtime"]="nvidia"
                docker_gpus=True
                container_options["device_requests"].append(docker.types.DeviceRequest(count=-1, capabilities=[['gpu']]))
            elif "gpus" in validated_container and type(validated_container["gpus"])==list:
                container_options["runtime"]="nvidia"
                docker_gpus=validated_container["gpus"]
                container_options["device_requests"].append(docker.types.DeviceRequest(
                    count=-1, 
                    capabilities=[['gpu']], 
                    device_ids=validated_container["gpus"]
                ))

            if "wireguard" in validated_container:
                wg_conf_dir = os.path.join(config.wireguard_config_folder, validated_container["name"])
                container_options["cap_add"].append('NET_ADMIN')
                container_options["cap_add"].append('SYS_MODULE')
                container_options["volumes"]["/lib/modules"] = {'bind': '/lib/modules', 'mode': 'ro'}
                container_options["volumes"][wg_conf_dir] = {'bind': '/config', 'mode': 'rw'}
            elif "allow_vpn" in validated_container:
                container_options["cap_add"].append('NET_ADMIN')
                container_options["cap_add"].append('SYS_MODULE')
                container_options["volumes"]["/lib/modules"] = {'bind': '/lib/modules', 'mode': 'ro'}

            if "limited_disk" in validated_container and type(validated_container["limited_disk"])==str:
                container_options["storage_opt"]={'size':validated_container["limited_disk"]}
            if "ports" in validated_container and type(validated_container["ports"])==list:
                for port in validated_container["ports"]:
                    if type(port)==str and ':' in port:
                        is_udp = True if "/udp" in port else False
                        port=port.replace('/udp','')
                        container_options["ports"][f"{port.split(':')[0]}/{'udp' if is_udp else 'tcp'}"]=int(port.split(':')[1])
            if "custom_entrypoint" in validated_container:
                entrypoint_file_name = f"{validated_container["name"]}.sh"
                entrypoint_full_path = os.path.join(config.entrypoints_folder, entrypoint_file_name)
                container_options["volumes"][entrypoint_full_path] = {'bind': '/etc/order_entrypoint.sh', 'mode': 'ro'}
                container_options["entrypoint"]='/etc/order_entrypoint.sh'
            elif "entrypoint_command" in validated_container and type(validated_container["entrypoint_command"])==str and len(validated_container["entrypoint_command"])>0:
                container_options["entrypoint"]=validated_container["entrypoint_command"]

            if not validated_container["name"] in created_container_names and image_ready:
                if config.creation_engine == "wrapper":
                    docker_cli_wrapper.create_container(container_options, ip=(validated_container["ip"] if "ip" in validated_container else None), docker_gpus=docker_gpus)
                else:
                    container = client.containers.create(**container_options)
                    if "ip" in validated_container:
                        client.networks.get(validated_container["network"] if "network" in validated_container else "clore-br0").connect(container, ipv4_address=validated_container["ip"])
                        client.networks.get("bridge").disconnect(container)
                    if not "paused" in validated_container:
                        container.start()
        except Exception as e:
            log.debug(f"Container creation issue | {e}")
            pass

    all_running_container_names = []
    all_stopped_container_names = []

    for container in all_containers:
        if type(container.name)==str:
            if container.status == "running":
                all_running_container_names.append(container.name)
            else:
                all_stopped_container_names.append(container.name)
            if container.name in needed_running_names and container.status != 'running':
                try:
                    attached_networks = container.attrs['NetworkSettings']['Networks']
                    if "bridge" in attached_networks.keys() or len(attached_networks.keys())==0: # Ip was not attached, remove container
                        container.stop()
                        container.remove()
                    else:
                        container.start()
                except Exception as e:
                    pass
            elif container.name in paused_names and container.status == 'running':
                try:
                    container.stop()
                except Exception as e:
                    pass
            elif container.name not in paused_names+needed_running_names+allowed_running_containers and container.status == 'running':
                try:
                    container.stop()
                    container.remove()
                except Exception as e:
                    pass
            elif container.name not in paused_names+needed_running_names+allowed_running_containers:
                try:
                    container.remove()
                except Exception as e:
                    pass

    return all_running_container_names, all_stopped_container_names
    #print(validated_containers)