pvc/NodeInstance.py

268 lines
12 KiB
Python
Raw Normal View History

2018-05-31 21:49:23 -04:00
#!/usr/bin/env python3
# NodeInstance.py - Class implementing a PVC node and run by pvcd
# Part of the Parallel Virtual Cluster (PVC) system
#
# Copyright (C) 2018 Joshua M. Boniface <joshua@boniface.me>
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program. If not, see <https://www.gnu.org/licenses/>.
#
###############################################################################
import os, sys, psutil, socket, time, libvirt, kazoo.client, threading, fencenode, ansiiprint
class NodeInstance():
2018-06-06 22:59:31 -04:00
# Initialization function
2018-06-08 12:19:48 -04:00
def __init__(self, name, t_node, s_domain, zk, config):
2018-05-31 21:49:23 -04:00
# Passed-in variables on creation
self.zk = zk
2018-06-08 12:19:48 -04:00
self.config = config
2018-05-31 21:49:23 -04:00
self.name = name
2018-06-01 01:00:55 -04:00
self.state = 'stop'
self.t_node = t_node
2018-06-04 16:34:41 -04:00
self.active_node_list = []
self.flushed_node_list = []
self.inactive_node_list = []
self.s_domain = s_domain
self.domain_list = []
2018-06-08 12:44:47 -04:00
self.ipmi_hostname = self.config['ipmi_hostname']
self.domains_count = 0
self.memused = 0
self.memfree = 0
# Zookeeper handlers for changed states
@zk.DataWatch('/nodes/{}/state'.format(self.name))
def watch_hypervisor_state(data, stat, event=""):
try:
self.state = data.decode('ascii')
except AttributeError:
self.state = 'stop'
2018-06-04 16:34:41 -04:00
if self.state == 'flush':
self.flush()
if self.state == 'unflush':
2018-06-04 16:34:41 -04:00
self.unflush()
@zk.DataWatch('/nodes/{}/memfree'.format(self.name))
def watch_hypervisor_memfree(data, stat, event=""):
try:
self.memfree = data.decode('ascii')
except AttributeError:
self.memfree = 0
@zk.DataWatch('/nodes/{}/runningdomains'.format(self.name))
def watch_hypervisor_runningdomains(data, stat, event=""):
try:
self.domain_list = data.decode('ascii').split()
except AttributeError:
self.domain_list = []
2018-05-31 21:49:23 -04:00
2018-05-31 22:31:20 -04:00
# Get value functions
2018-05-31 23:41:52 -04:00
def getfreemem(self):
2018-05-31 22:31:20 -04:00
return self.memfree
2018-05-31 23:41:52 -04:00
def getcpuload(self):
2018-05-31 22:31:20 -04:00
return self.cpuload
2018-05-31 23:41:52 -04:00
def getname(self):
2018-05-31 22:31:20 -04:00
return self.name
2018-05-31 23:41:52 -04:00
def getstate(self):
2018-05-31 23:40:21 -04:00
return self.state
def getdomainlist(self):
return self.domain_list
2018-05-31 23:01:22 -04:00
# Update value functions
def updatenodelist(self, t_node):
self.t_node = t_node
2018-05-31 23:01:22 -04:00
def updatedomainlist(self, s_domain):
self.s_domain = s_domain
2018-05-31 22:31:20 -04:00
# Flush all VMs on the host
2018-06-04 16:34:41 -04:00
def flush(self):
2018-06-06 22:59:31 -04:00
ansiiprint.echo('Flushing node "{}" of running VMs'.format(self.name), '', 'i')
for dom_uuid in self.domain_list:
most_memfree = 0
2018-06-06 21:31:20 -04:00
target_hypervisor = None
2018-06-06 20:49:21 -04:00
hypervisor_list = self.zk.get_children('/nodes')
2018-06-06 21:28:02 -04:00
current_hypervisor = self.zk.get('/domains/{}/hypervisor'.format(dom_uuid))[0].decode('ascii')
for hypervisor in hypervisor_list:
2018-06-06 20:49:21 -04:00
state = self.zk.get('/nodes/{}/state'.format(hypervisor))[0].decode('ascii')
if state != 'start' or hypervisor == current_hypervisor:
2018-06-04 02:22:59 -04:00
continue
2018-06-06 20:49:21 -04:00
memfree = int(self.zk.get('/nodes/{}/memfree'.format(hypervisor))[0].decode('ascii'))
if memfree > most_memfree:
most_memfree = memfree
target_hypervisor = hypervisor
2018-06-06 21:28:58 -04:00
if target_hypervisor == None:
2018-06-06 22:59:31 -04:00
ansiiprint.echo('Failed to find migration target for VM "{}"; shutting down'.format(dom_uuid), '', 'e')
2018-06-04 02:22:59 -04:00
transaction = self.zk.transaction()
transaction.set_data('/domains/{}/state'.format(dom_uuid), 'shutdown'.encode('ascii'))
2018-06-04 02:22:59 -04:00
transaction.commit()
else:
2018-06-06 22:59:31 -04:00
ansiiprint.echo('Migrating VM "{}" to hypervisor "{}"'.format(dom_uuid, '', target_hypervisor), 'i')
2018-06-04 02:22:59 -04:00
transaction = self.zk.transaction()
2018-06-06 21:35:32 -04:00
transaction.set_data('/domains/{}/state'.format(dom_uuid), 'migrate'.encode('ascii'))
transaction.set_data('/domains/{}/hypervisor'.format(dom_uuid), target_hypervisor.encode('ascii'))
transaction.set_data('/domains/{}/lasthypervisor'.format(dom_uuid), current_hypervisor.encode('ascii'))
result = transaction.commit()
2018-06-04 02:22:59 -04:00
# Wait 1s between migrations
time.sleep(1)
2018-05-31 22:31:20 -04:00
2018-06-04 03:09:51 -04:00
def unflush(self):
2018-06-06 22:59:31 -04:00
ansiiprint.echo('Restoring node {} to active service.'.format(self.name), '', 'i')
self.zk.set('/nodes/{}/state'.format(self.name), 'start'.encode('ascii'))
for dom_uuid in self.s_domain:
last_hypervisor = self.zk.get('/domains/{}/lasthypervisor'.format(dom_uuid))[0].decode('ascii')
if last_hypervisor != self.name:
continue
2018-06-06 22:59:31 -04:00
ansiiprint.echo('Setting unmigration for VM "{}"'.format(dom_uuid), '', 'i')
transaction = self.zk.transaction()
transaction.set_data('/domains/{}/state'.format(dom_uuid), 'migrate'.encode('ascii'))
transaction.set_data('/domains/{}/hypervisor'.format(dom_uuid), self.name.encode('ascii'))
transaction.set_data('/domains/{}/lasthypervisor'.format(dom_uuid), ''.encode('ascii'))
result = transaction.commit()
2018-06-04 03:09:51 -04:00
# Wait 1s between migrations
time.sleep(1)
2018-06-04 03:09:51 -04:00
def update_zookeeper(self):
2018-05-31 21:49:23 -04:00
# Connect to libvirt
libvirt_name = "qemu:///system"
conn = libvirt.open(libvirt_name)
print(conn.listDomainsID())
print(psutil.virtual_memory())
2018-05-31 21:49:23 -04:00
if conn == None:
2018-06-06 22:59:31 -04:00
ansiiprint.echo('Failed to open connection to "{}"'.format(libvirt_name), '', 'e')
return
2018-05-31 21:49:23 -04:00
2018-06-04 02:34:03 -04:00
# Get past state and update if needed
2018-06-06 20:42:33 -04:00
past_state = self.zk.get('/nodes/{}/state'.format(self.name))[0].decode('ascii')
2018-06-04 02:34:03 -04:00
if past_state != 'flush':
self.state = 'start'
2018-06-06 20:42:33 -04:00
self.zk.set('/nodes/{}/state'.format(self.name), 'start'.encode('ascii'))
2018-06-04 02:34:03 -04:00
else:
self.state = 'flush'
# Toggle state management of all VMs and remove any non-running VMs
for domain, instance in self.s_domain.items():
if instance.inshutdown == False and domain in self.domain_list:
instance.manage_vm_state()
if instance.dom == None:
try:
self.domain_list.remove(domain)
except:
pass
else:
try:
state = instance.dom.state()[0]
except:
state = libvirt.VIR_DOMAIN_NOSTATE
if state != libvirt.VIR_DOMAIN_RUNNING:
2018-06-04 01:26:23 -04:00
try:
self.domain_list.remove(domain)
except:
pass
# Set our information in zookeeper
self.name = conn.getHostname()
self.cpucount = conn.getCPUMap()[0]
self.memused = int(psutil.virtual_memory().used / 1024 / 1024)
self.memfree = int(psutil.virtual_memory().free / 1024 / 1024)
self.cpuload = os.getloadavg()[0]
self.domains_count = len(conn.listDomainsID())
keepalive_time = int(time.time())
try:
2018-06-06 20:42:33 -04:00
self.zk.set('/nodes/{}/cpucount'.format(self.name), str(self.cpucount).encode('ascii'))
self.zk.set('/nodes/{}/memused'.format(self.name), str(self.memused).encode('ascii'))
2018-06-06 20:42:33 -04:00
self.zk.set('/nodes/{}/memfree'.format(self.name), str(self.memfree).encode('ascii'))
self.zk.set('/nodes/{}/cpuload'.format(self.name), str(self.cpuload).encode('ascii'))
self.zk.set('/nodes/{}/runningdomains'.format(self.name), ' '.join(self.domain_list).encode('ascii'))
self.zk.set('/nodes/{}/domainscount'.format(self.name), str(self.domains_count).encode('ascii'))
2018-06-06 20:42:33 -04:00
self.zk.set('/nodes/{}/keepalive'.format(self.name), str(keepalive_time).encode('ascii'))
except:
return
2018-06-06 15:16:39 -04:00
# Close the Libvirt connection
conn.close()
# Display node information to the terminal
ansiiprint.echo('{}{} keepalive{}'.format(ansiiprint.purple(), self.name, ansiiprint.end()), '', 't')
ansiiprint.echo('{0}CPUs:{1} {2} {0}Free memory [MiB]:{1} {3} {0}Used memory [MiB]:{1} {4} {0}Load:{1} {5}'.format(ansiiprint.bold(), ansiiprint.end(), self.cpucount, self.memfree, self.memused, self.cpuload), '', 'c')
2018-06-11 01:51:19 -04:00
ansiiprint.echo('{}Active domains:{} {} [{}]'.format(ansiiprint.bold(), ansiiprint.end(), self.domains_count, ','.join(self.domain_list)), '', 'c')
# Update our local node lists
for node_name in self.t_node:
2018-06-01 01:00:55 -04:00
try:
node_state = self.zk.get('/nodes/{}/state'.format(node_name))[0].decode('ascii')
node_keepalive = int(self.zk.get('/nodes/{}/keepalive'.format(node_name))[0].decode('ascii'))
2018-06-01 01:00:55 -04:00
except:
node_state = 'unknown'
node_keepalive = 0
# Handle deadtime and fencng if needed
2018-06-08 12:44:47 -04:00
# (A node is considered dead when its keepalive timer is >6*keepalive_interval seconds
# out-of-date while in 'start' state)
node_deadtime = int(time.time()) - ( int(self.config['keepalive_interval']) * 6 )
if node_keepalive < node_deadtime and node_state == 'start':
2018-06-06 22:59:31 -04:00
ansiiprint.echo('Node {} is dead - performing fence operation in 3 seconds'.format(node_name), '', 'w')
self.zk.set('/nodes/{}/state'.format(node_name), 'dead'.encode('ascii'))
fence_thread = threading.Thread(target=fencenode.fence, args=(node_name, self.zk), kwargs={})
fence_thread.start()
# Update the arrays
if node_state == 'start' and node_name not in self.active_node_list:
self.active_node_list.append(node_name)
2018-06-06 00:49:34 -04:00
try:
self.flushed_node_list.remove(node_name)
except ValueError:
pass
try:
self.inactive_node_list.remove(node_name)
except ValueError:
pass
if node_state == 'flush' and node_name not in self.flushed_node_list:
self.flushed_node_list.append(node_name)
try:
self.active_node_list.remove(node_name)
except ValueError:
pass
try:
self.inactive_node_list.remove(node_name)
except ValueError:
pass
if node_state != 'start' and node_state != 'flush' and node_name not in self.inactive_node_list:
self.inactive_node_list.append(node_name)
try:
self.active_node_list.remove(node_name)
except ValueError:
pass
try:
self.flushed_node_list.remove(node_name)
except ValueError:
pass
# Display cluster information to the terminal
ansiiprint.echo('{}Cluster status{}'.format(ansiiprint.purple(), ansiiprint.end()), '', 't')
ansiiprint.echo('{}Active nodes:{} {}'.format(ansiiprint.bold(), ansiiprint.end(), ' '.join(self.active_node_list)), '', 'c')
ansiiprint.echo('{}Flushed nodes:{} {}'.format(ansiiprint.bold(), ansiiprint.end(), ' '.join(self.flushed_node_list)), '', 'c')
ansiiprint.echo('{}Inactive nodes:{} {}'.format(ansiiprint.bold(), ansiiprint.end(), ' '.join(self.inactive_node_list)), '', 'c')