mravi/uncloud-mravi
1
0
Fork 0
forked from uncloud/uncloud
Code Issues Pull requests Projects Releases Wiki Activity

1. mp.set_start_method('spawn') commented out from scripts/uncloud

Browse source 
2. uncloud.shared moved under uncloud.common
3. Refactoring in etcd_wrapper e.g timeout mechanism removed and few other things
4. uncloud-{scheduler,host} now better handle etcd events in their block state (waiting for requests to come)
This commit is contained in:
ahmadbilalkhalid 2020-01-09 00:40:05 +05:00
commit 48efcdf08c
17 changed files with 136 additions and 173 deletions
Download patch file Download diff file Expand all files Collapse all files

76
uncloud/scheduler/main.py
View file

@ -6,59 +6,51 @@
import argparse
from uncloud.common.request import RequestEntry, RequestType
from uncloud.shared import shared
from uncloud.common.settings import settings
from .helper import (dead_host_mitigation, dead_host_detection, assign_host, NoSuitableHostFound)
from . import logger
from uncloud.common.request import RequestEntry, RequestType
from uncloud.common.shared import shared
from uncloud.scheduler import logger
from uncloud.scheduler.helper import (dead_host_mitigation, dead_host_detection,
assign_host, NoSuitableHostFound)
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":
# The below while True is neccessary for gracefully handling leadership transfer and temporary
# unavailability in etcd. Why does it work? It works because the get_prefix,watch_prefix return
# iter([]) that is iterator of empty list on exception (that occur due to above mentioned reasons)
# which ends the loop immediately. So, having it inside infinite loop we try again and again to
# get prefix until either success or deamon death comes.
while True:
for request_iterator in [
shared.etcd_client.get_prefix(settings['etcd']['request_prefix'], value_in_json=True,
raise_exception=False),
shared.etcd_client.watch_prefix(settings['etcd']['request_prefix'], value_in_json=True,
raise_exception=False),
]:
for request_event in request_iterator:
dead_host_mitigation(dead_host_detection())
request_entry = RequestEntry(request_event)
# 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)
if request_entry.type == RequestType.ScheduleVM:
logger.debug('%s, %s', request_entry.key, request_entry.value)
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
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
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)
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....")
logger.info('No Resource Left. Emailing admin....')
if __name__ == "__main__":
if __name__ == '__main__':
main()