uncloud/ucloud/host/main.py

import argparse
import multiprocessing as mp
import time

from etcd3_wrapper import Etcd3Wrapper

from ucloud.common.request import RequestEntry, RequestType
from ucloud.config import (vm_pool, request_pool,
                    etcd_client, running_vms,
                    etcd_wrapper_args, etcd_wrapper_kwargs,
                    HostPool, env_vars)

from .helper import find_free_port
from . import virtualmachine
from ucloud.host import logger


def update_heartbeat(hostname):
    """Update Last HeartBeat Time for :param hostname: in etcd"""
    client = Etcd3Wrapper(*etcd_wrapper_args, **etcd_wrapper_kwargs)
    host_pool = HostPool(client, env_vars.get('HOST_PREFIX'))
    this_host = next(filter(lambda h: h.hostname == hostname, host_pool.hosts), None)

    while True:
        this_host.update_heartbeat()
        host_pool.put(this_host)
        time.sleep(10)


def maintenance(host):
    # To capture vm running according to running_vms list

    # This is to capture successful migration of a VM.
    # Suppose, this host is running "vm1" and user initiated
    # request to migrate this "vm1" to some other host. On,
    # successful migration the destination host would set
    # the vm hostname to itself. Thus, we are checking
    # whether this host vm is successfully migrated. If yes
    # then we shutdown "vm1" on this host.

    to_be_removed = []
    for running_vm in running_vms:
        with vm_pool.get_put(running_vm.key) as vm_entry:
            if vm_entry.hostname != host.key and not vm_entry.in_migration:
                running_vm.handle.shutdown()
                logger.info("VM migration not completed successfully.")
                to_be_removed.append(running_vm)

    for r in to_be_removed:
        running_vms.remove(r)

    # To check vm running according to etcd entries
    alleged_running_vms = vm_pool.by_status("RUNNING", vm_pool.by_host(host.key))

    for vm_entry in alleged_running_vms:
        _vm = virtualmachine.get_vm(running_vms, vm_entry.key)
        # Whether, the allegedly running vm is in our
        # running_vms list or not if it is said to be
        # running on this host but it is not then we
        # need to shut it down

        # This is to capture poweroff/shutdown of a VM
        # initiated by user inside VM. OR crash of VM by some
        # user running process
        if (_vm and not _vm.handle.is_running()) or not _vm:
            logger.debug("_vm = %s, is_running() = %s" % (_vm, _vm.handle.is_running()))
            vm_entry.add_log("""{} is not running but is said to be running.
                                So, shutting it down and declare it killed""".format(vm_entry.key))
            vm_entry.declare_killed()
            vm_pool.put(vm_entry)
            if _vm:
                running_vms.remove(_vm)


def main(hostname):
    heartbeat_updating_process = mp.Process(target=update_heartbeat, args=(hostname,))

    host_pool = HostPool(etcd_client, env_vars.get('HOST_PREFIX'))
    host = next(filter(lambda h: h.hostname == hostname, host_pool.hosts), None)
    assert host is not None, "No such host with name = {}".format(hostname)

    try:
        heartbeat_updating_process.start()
    except Exception as e:
        logger.info("No Need To Go Further. Our heartbeat updating mechanism is not working")
        logger.exception(e)
        exit(-1)

    logger.info("%s Session Started %s", '*' * 5, '*' * 5)

    # It is seen that under heavy load, timeout event doesn't come
    # in a predictive manner (which is intentional because we give
    # higher priority to customer's requests) which delays heart
    # beat update which in turn misunderstood by scheduler that the
    # host is dead when it is actually alive. So, to ensure that we
    # update the heart beat in a predictive manner we start Heart
    # beat updating mechanism in separated thread

    for events_iterator in [
        etcd_client.get_prefix(env_vars.get('REQUEST_PREFIX'), value_in_json=True),
        etcd_client.watch_prefix(env_vars.get('REQUEST_PREFIX'), timeout=10, value_in_json=True),
    ]:
        for request_event in events_iterator:
            request_event = RequestEntry(request_event)

            if request_event.type == "TIMEOUT":
                maintenance(host)
                continue

            # If the event is directed toward me OR I am destination of a InitVMMigration
            if request_event.hostname == host.key or request_event.destination == host.key:
                logger.debug("VM Request: %s", request_event)

                request_pool.client.client.delete(request_event.key)
                vm_entry = vm_pool.get(request_event.uuid)

                if vm_entry:
                    if request_event.type == RequestType.StartVM:
                        virtualmachine.start(vm_entry)

                    elif request_event.type == RequestType.StopVM:
                        virtualmachine.stop(vm_entry)

                    elif request_event.type == RequestType.DeleteVM:
                        virtualmachine.delete(vm_entry)

                    elif request_event.type == RequestType.InitVMMigration:
                        virtualmachine.start(vm_entry, host.key, find_free_port())

                    elif request_event.type == RequestType.TransferVM:
                        virtualmachine.transfer(request_event)
                else:
                    logger.info("VM Entry missing")

                logger.info("Running VMs %s", running_vms)


if __name__ == "__main__":
    argparser = argparse.ArgumentParser()
    argparser.add_argument("hostname", help="Name of this host. e.g /v1/host/1")
    args = argparser.parse_args()
    mp.set_start_method('spawn')
    main(args.hostname)