X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=sliver_libvirt.py;h=e898c4baa0fd3007ca5e7dec7160938a3be3ca44;hb=570d234b7d7b40416b296de71225f28c826ee991;hp=f4998b0e3edfb5ef628820b06b6199c81b9b0325;hpb=1459f786a35efaaf2869af187acdb609ddd1cffb;p=nodemanager.git
diff --git a/sliver_libvirt.py b/sliver_libvirt.py
index f4998b0..e898c4b 100644
--- a/sliver_libvirt.py
+++ b/sliver_libvirt.py
@@ -2,8 +2,10 @@
import sys
import os, os.path
+import re
import subprocess
import pprint
+import random
import libvirt
@@ -22,6 +24,13 @@ STATES = {
libvirt.VIR_DOMAIN_CRASHED: 'crashed',
}
+REASONS = {
+ libvirt.VIR_CONNECT_CLOSE_REASON_ERROR: 'Misc I/O error',
+ libvirt.VIR_CONNECT_CLOSE_REASON_EOF: 'End-of-file from server',
+ libvirt.VIR_CONNECT_CLOSE_REASON_KEEPALIVE: 'Keepalive timer triggered',
+ libvirt.VIR_CONNECT_CLOSE_REASON_CLIENT: 'Client requested it',
+}
+
connections = dict()
# Common Libvirt code
@@ -32,21 +41,37 @@ class Sliver_Libvirt(Account):
@staticmethod
def getConnection(sliver_type):
- # TODO: error checking
- # vtype is of the form sliver.[LXC/QEMU] we need to lower case to lxc/qemu
+ """
+ returns a connection to the underlying libvirt service
+ a single connection is created and shared among slivers
+ this call ensures the connection is alive
+ and will reconnect if it appears to be necessary
+ """
+ # sliver_type comes from rec['type'] and is of the form sliver.{LXC,QEMU}
+ # so we need to lower case to lxc/qemu
vtype = sliver_type.split('.')[1].lower()
- uri = vtype + '://'
- return connections.setdefault(uri, libvirt.open(uri))
-
- @staticmethod
- def debuginfo(dom):
- ''' Helper method to get a "nice" output of the info struct for debug'''
- [state, maxmem, mem, ncpu, cputime] = dom.info()
- return '%s is %s, maxmem = %s, mem = %s, ncpu = %s, cputime = %s' % (dom.name(), STATES.get(state, state), maxmem, mem, ncpu, cputime)
+ uri = vtype + ':///'
+ if uri not in connections:
+ # create connection
+ conn = libvirt.open(uri)
+ connections[uri] = conn
+ return conn
+ else:
+ # connection already available : check for health
+ conn = connections[uri]
+ # see if a reconnection is needed
+ try:
+ numDomains = conn.numOfDomains()
+ except:
+ logger.log("libvirt connection to {} looks broken - reconnecting".format(uri))
+ conn = libvirt.open(uri)
+ # if this fails then an expection is thrown outside of this function
+ numDomains = conn.numOfDomains()
+ return conn
def __init__(self, rec):
self.name = rec['name']
- logger.verbose ('sliver_libvirt: %s init'%(self.name))
+ logger.verbose ('sliver_libvirt: {} init'.format(self.name))
# Assume the directory with the image and config files
# are in place
@@ -62,70 +87,114 @@ class Sliver_Libvirt(Account):
try:
dom = self.conn.lookupByName(self.name)
except:
- logger.log('sliver_libvirt: Domain %s does not exist. ' \
- 'Will try to create it again.' % (self.name))
+ logger.log('sliver_libvirt: Domain {} does not exist. ' \
+ 'Will try to create it again.'.format(self.name))
self.__class__.create(rec['name'], rec)
dom = self.conn.lookupByName(self.name)
self.dom = dom
+ @staticmethod
+ def dom_details (dom):
+ output = ""
+ output += " id={} - OSType={}".format(dom.ID(), dom.OSType())
+ # calling state() seems to be working fine
+ (state, reason) = dom.state()
+ output += " state={}, reason={}".format(STATES.get(state, state),
+ REASONS.get(reason, reason))
+ try:
+ # try to use info() - this however does not work for some reason on f20
+ # info cannot get info operation failed: Cannot read cputime for domain
+ [state, maxmem, mem, ncpu, cputime] = dom.info()
+ output += " [info: state={}, maxmem = {}, mem = {}, ncpu = {}, cputime = {}]"\
+ .format(STATES.get(state, state), maxmem, mem, ncpu, cputime)
+ except:
+ # too bad but libvirt.py prints out stuff on stdout when this fails, don't know how to get rid of that..
+ output += " [info: not available]"
+ return output
+
+ def __repr__(self):
+ ''' Helper method to get a "nice" output of the domain struct for debug purposes'''
+ output = "Domain {}".format(self.name)
+ dom = self.dom
+ if dom is None:
+ output += " [no attached dom ?!?]"
+ else:
+ output += Sliver_Libvirt.dom_details (dom)
+ return output
+
+ # Thierry : I am not quite sure if /etc/libvirt/lxc/<>.xml holds a reliably up-to-date
+ # copy of the sliver XML config; I feel like issuing a virsh dumpxml first might be safer
+ def repair_veth(self):
+ # See workaround email, 2-14-2014, "libvirt 1.2.1 rollout"
+ xmlfilename = "/etc/libvirt/lxc/{}.xml".format(self.name)
+ with open(xmlfilename) as xmlfile:
+ xml = xmlfile.read()
+ veths = re.findall("", xml)
+ veths = [x[13:-3] for x in veths]
+ for veth in veths:
+ command = ["ip", "link", "delete", veth]
+ logger.log_call(command)
+
+ logger.log("trying to redefine the VM")
+ command = [ "virsh", "define", xmlfilename ]
+ logger.log_call(command)
+
def start(self, delay=0):
- ''' Just start the sliver '''
- logger.verbose('sliver_libvirt: %s start'%(self.name))
+ '''Just start the sliver'''
+ logger.verbose('sliver_libvirt: {} start'.format(self.name))
# Check if it's running to avoid throwing an exception if the
- # domain was already running, create actually means start
+ # domain was already running
if not self.is_running():
- self.dom.create()
+ try:
+ # create actually means start
+ self.dom.create()
+ except Exception, e:
+ # XXX smbaker: attempt to resolve slivers that are stuck in
+ # "failed to allocate free veth".
+ if "ailed to allocate free veth" in str(e):
+ logger.log("failed to allocate free veth on {}".format(self.name))
+ self.repair_veth()
+ logger.log("trying dom.create again")
+ self.dom.create()
+ else:
+ raise
else:
- logger.verbose('sliver_libvirt: sliver %s already started'%(self.name))
+ logger.verbose('sliver_libvirt: sliver {} already started'.format(self.name))
# After the VM is started... we can play with the virtual interface
# Create the ebtables rule to mark the packets going out from the virtual
# interface to the actual device so the filter canmatch against the mark
- bwlimit.ebtables("-A INPUT -i veth%d -j mark --set-mark %d" % \
- (self.xid, self.xid))
+ bwlimit.ebtables("-A INPUT -i veth{} -j mark --set-mark {}"
+ .format(self.xid, self.xid))
+ ### this is confusing, because it seems it is not used in fact
def stop(self):
- logger.verbose('sliver_libvirt: %s stop'%(self.name))
+ logger.verbose('sliver_libvirt: {} stop'.format(self.name))
# Remove the ebtables rule before stopping
- bwlimit.ebtables("-D INPUT -i veth%d -j mark --set-mark %d" % \
- (self.xid, self.xid))
+ bwlimit.ebtables("-D INPUT -i veth{} -j mark --set-mark {}"
+ .format(self.xid, self.xid))
try:
self.dom.destroy()
except:
- logger.verbose('sliver_libvirt: Domain %s not running ' \
- 'UNEXPECTED: %s'%(self.name, sys.exc_info()[1]))
- print 'sliver_libvirt: Domain %s not running ' \
- 'UNEXPECTED: %s'%(self.name, sys.exc_info()[1])
+ logger.log_exc("in sliver_libvirt.stop", name=self.name)
def is_running(self):
''' Return True if the domain is running '''
- logger.verbose('sliver_libvirt: %s is_running'%self.name)
- try:
- [state, _, _, _, _] = self.dom.info()
- if state == libvirt.VIR_DOMAIN_RUNNING:
- logger.verbose('sliver_libvirt: %s is RUNNING'%self.name)
- return True
- else:
- info = debuginfo(self.dom)
- logger.verbose('sliver_libvirt: %s is ' \
- 'NOT RUNNING...\n%s'%(self.name, info))
- return False
- except:
- logger.verbose('sliver_libvirt: UNEXPECTED ERROR in ' \
- '%s: %s'%(self.name, sys.exc_info()[1]))
- print 'sliver_libvirt: UNEXPECTED ERROR in ' \
- '%s: %s'%(self.name, sys.exc_info()[1])
- return False
+ (state, _) = self.dom.state()
+ result = (state == libvirt.VIR_DOMAIN_RUNNING)
+ logger.verbose('sliver_libvirt.is_running: {} => {}'
+ .format(self, result))
+ return result
def configure(self, rec):
#sliver.[LXC/QEMU] tolower case
#sliver_type = rec['type'].split('.')[1].lower()
- #BASE_DIR = '/cgroup/libvirt/%s/%s/'%(sliver_type, self.name)
+ #BASE_DIR = '/cgroup/libvirt/{}/{}/'.format(sliver_type, self.name)
# Disk allocation
# No way through cgroups... figure out how to do that with user/dir quotas.
@@ -136,43 +205,55 @@ class Sliver_Libvirt(Account):
# Btrfs support quota per volumes
if rec.has_key("rspec") and rec["rspec"].has_key("tags"):
- tags = rec["rspec"]["tags"]
- # It will depend on the FS selection
- if tags.has_key('disk_max'):
- disk_max = tags['disk_max']
- if disk_max == 0:
- # unlimited
- pass
- else:
- # limit to certain number
- pass
-
- # Memory allocation
- if tags.has_key('memlock_hard'):
- mem = str(int(tags['memlock_hard']) * 1024) # hard limit in bytes
- cgroups.write(self.name, 'memory.limit_in_bytes', mem, subsystem="memory")
- if tags.has_key('memlock_soft'):
- mem = str(int(tags['memlock_soft']) * 1024) # soft limit in bytes
- cgroups.write(self.name, 'memory.soft_limit_in_bytes', mem, subsystem="memory")
-
- # CPU allocation
- # Only cpu_shares until figure out how to provide limits and guarantees
- # (RT_SCHED?)
- if tags.has_key('cpu_share'):
- cpu_share = tags['cpu_share']
- cgroups.write(self.name, 'cpu.shares', cpu_share)
+ if cgroups.get_cgroup_path(self.name) == None:
+ # If configure is called before start, then the cgroups won't exist
+ # yet. NM will eventually re-run configure on the next iteration.
+ # TODO: Add a post-start configure, and move this stuff there
+ logger.log("Configure: postponing tag check on {} as cgroups are not yet populated"
+ .format(self.name))
+ else:
+ tags = rec["rspec"]["tags"]
+ # It will depend on the FS selection
+ if tags.has_key('disk_max'):
+ disk_max = tags['disk_max']
+ if disk_max == 0:
+ # unlimited
+ pass
+ else:
+ # limit to certain number
+ pass
+
+ # Memory allocation
+ if tags.has_key('memlock_hard'):
+ mem = str(int(tags['memlock_hard']) * 1024) # hard limit in bytes
+ cgroups.write(self.name, 'memory.limit_in_bytes', mem, subsystem="memory")
+ if tags.has_key('memlock_soft'):
+ mem = str(int(tags['memlock_soft']) * 1024) # soft limit in bytes
+ cgroups.write(self.name, 'memory.soft_limit_in_bytes', mem, subsystem="memory")
+
+ # CPU allocation
+ # Only cpu_shares until figure out how to provide limits and guarantees
+ # (RT_SCHED?)
+ if tags.has_key('cpu_share'):
+ cpu_share = tags['cpu_share']
+ cgroups.write(self.name, 'cpu.shares', cpu_share)
# Call the upper configure method (ssh keys...)
Account.configure(self, rec)
+ @staticmethod
+ def get_unique_vif():
+ return 'veth{}'.format(random.getrandbits(32))
+
# A placeholder until we get true VirtualInterface objects
@staticmethod
def get_interfaces_xml(rec):
xml = """
+
-"""
+""".format(Sliver_Libvirt.get_unique_vif())
try:
tags = rec['rspec']['tags']
if 'interface' in tags:
@@ -183,28 +264,31 @@ class Sliver_Libvirt(Account):
tag_xml = ""
for interface in interfaces:
if 'vlan' in interface:
- vlanxml = "" % interface['vlan']
+ vlanxml = "".format(interface['vlan'])
else:
vlanxml = ""
if 'bridge' in interface:
tag_xml = tag_xml + """
-
- %s
+
+ {}
+
- """ % (interface['bridge'], vlanxml)
+ """.format(interface['bridge'], vlanxml, Sliver_Libvirt.get_unique_vif())
else:
tag_xml = tag_xml + """
+
- """
+ """.format(Sliver_Libvirt.get_unique_vif())
+
xml = tag_xml
- logger.log('sliver_libvirty.py: interface XML is: %s' % xml)
+ logger.log('sliver_libvirty.py: interface XML is: {}'.format(xml))
except:
- logger.log('sliver_libvirt.py: ERROR parsing "interface" tag for slice %s' % rec['name'])
- logger.log('sliver_libvirt.py: tag value: %s' % tags['interface'])
+ logger.log('sliver_libvirt.py: ERROR parsing "interface" tag for slice {}'.format(rec['name']))
+ logger.log('sliver_libvirt.py: tag value: {}'.format(tags['interface']))
return xml