uncloud-mravi/uncloud/scheduler/main.py

64 lines
2.4 KiB
Python
Executable file

# TODO
# 1. send an email to an email address defined by env['admin-email']
# if resources are finished
# 2. Introduce a status endpoint of the scheduler -
# maybe expose a prometheus compatible output
import argparse
from uncloud.common.request import RequestEntry, RequestType
from uncloud.shared import shared
from uncloud.settings import settings
from .helper import (dead_host_mitigation, dead_host_detection, assign_host, NoSuitableHostFound)
from . import logger
arg_parser = argparse.ArgumentParser('scheduler', add_help=False)
def main(debug=False):
for request_iterator in [
shared.etcd_client.get_prefix(
settings["etcd"]["request_prefix"], value_in_json=True
),
shared.etcd_client.watch_prefix(
settings["etcd"]["request_prefix"],
timeout=5,
value_in_json=True,
),
]:
for request_event in request_iterator:
request_entry = RequestEntry(request_event)
# Never Run time critical mechanism inside timeout
# mechanism because timeout mechanism only comes
# when no other event is happening. It means under
# heavy load there would not be a timeout event.
if request_entry.type == "TIMEOUT":
# Detect hosts that are dead and set their status
# to "DEAD", and their VMs' status to "KILLED"
dead_hosts = dead_host_detection()
if dead_hosts:
logger.debug("Dead hosts: %s", dead_hosts)
dead_host_mitigation(dead_hosts)
elif request_entry.type == RequestType.ScheduleVM:
logger.debug("%s, %s", request_entry.key, request_entry.value)
vm_entry = shared.vm_pool.get(request_entry.uuid)
if vm_entry is None:
logger.info("Trying to act on {} but it is deleted".format(request_entry.uuid))
continue
shared.etcd_client.client.delete(request_entry.key) # consume Request
try:
assign_host(vm_entry)
except NoSuitableHostFound:
vm_entry.add_log("Can't schedule VM. No Resource Left.")
shared.vm_pool.put(vm_entry)
logger.info("No Resource Left. Emailing admin....")
if __name__ == "__main__":
main()