uncloud/ucloud/scheduler/helper.py

139 lines
4 KiB
Python
Raw Normal View History

from collections import Counter
from functools import reduce
import bitmath
from ucloud.common.host import HostStatus
from ucloud.common.request import RequestEntry, RequestType
from ucloud.common.vm import VMStatus
from ucloud.shared import shared
from ucloud.settings import settings
def accumulated_specs(vms_specs):
if not vms_specs:
return {}
return reduce((lambda x, y: Counter(x) + Counter(y)), vms_specs)
def remaining_resources(host_specs, vms_specs):
# Return remaining resources host_specs - vms
_vms_specs = Counter(vms_specs)
_remaining = Counter(host_specs)
for component in _vms_specs:
if isinstance(_vms_specs[component], str):
_vms_specs[component] = int(
bitmath.parse_string_unsafe(
_vms_specs[component]
).to_MB()
)
elif isinstance(_vms_specs[component], list):
_vms_specs[component] = map(
lambda x: int(bitmath.parse_string_unsafe(x).to_MB()),
_vms_specs[component],
)
_vms_specs[component] = reduce(
lambda x, y: x + y, _vms_specs[component], 0
)
for component in _remaining:
if isinstance(_remaining[component], str):
_remaining[component] = int(
bitmath.parse_string_unsafe(
_remaining[component]
).to_MB()
)
elif isinstance(_remaining[component], list):
_remaining[component] = map(
lambda x: int(bitmath.parse_string_unsafe(x).to_MB()),
_remaining[component],
)
_remaining[component] = reduce(
lambda x, y: x + y, _remaining[component], 0
)
_remaining.subtract(_vms_specs)
return _remaining
class NoSuitableHostFound(Exception):
"""Exception when no host found that can host a VM."""
def get_suitable_host(vm_specs, hosts=None):
if hosts is None:
hosts = shared.host_pool.by_status(HostStatus.alive)
for host in hosts:
# Filter them by host_name
vms = shared.vm_pool.by_host(host.key)
# Filter them by status
vms = shared.vm_pool.by_status(VMStatus.running, vms)
running_vms_specs = [vm.specs for vm in vms]
# Accumulate all of their combined specs
running_vms_accumulated_specs = accumulated_specs(
running_vms_specs
)
# Find out remaining resources after
# host_specs - already running vm_specs
remaining = remaining_resources(
host.specs, running_vms_accumulated_specs
)
# Find out remaining - new_vm_specs
remaining = remaining_resources(remaining, vm_specs)
if all(map(lambda x: x >= 0, remaining.values())):
return host.key
raise NoSuitableHostFound
def dead_host_detection():
# Bring out your dead! - Monty Python and the Holy Grail
hosts = shared.host_pool.by_status(HostStatus.alive)
dead_hosts_keys = []
for host in hosts:
# Only check those who claims to be alive
if host.status == HostStatus.alive:
if not host.is_alive():
dead_hosts_keys.append(host.key)
return dead_hosts_keys
def dead_host_mitigation(dead_hosts_keys):
for host_key in dead_hosts_keys:
host = shared.host_pool.get(host_key)
host.declare_dead()
vms_hosted_on_dead_host = shared.vm_pool.by_host(host_key)
for vm in vms_hosted_on_dead_host:
vm.status = "UNKNOWN"
shared.vm_pool.put(vm)
shared.host_pool.put(host)
def assign_host(vm):
vm.hostname = get_suitable_host(vm.specs)
shared.vm_pool.put(vm)
r = RequestEntry.from_scratch(
type=RequestType.StartVM,
uuid=vm.uuid,
hostname=vm.hostname,
request_prefix=settings["etcd"]["request_prefix"],
)
shared.request_pool.put(r)
vm.log.append("VM scheduled for starting")
return vm.hostname