X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=sliver_libvirt.py;h=8189b192c7f67ae6d96c6db6f77d0c1d364b3e4a;hb=e57432c1dfdfeaa52cc32799e2abbc34b7704ce9;hp=753b0c3a297dfb112d38801cec8948f6267750d6;hpb=34bbf772786e0e674d2b83ddfb079cb060830ff1;p=nodemanager.git
diff --git a/sliver_libvirt.py b/sliver_libvirt.py
index 753b0c3..8189b19 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
@@ -38,12 +47,6 @@ class Sliver_Libvirt(Account):
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)
-
def __init__(self, rec):
self.name = rec['name']
logger.verbose ('sliver_libvirt: %s init'%(self.name))
@@ -68,14 +71,68 @@ class Sliver_Libvirt(Account):
dom = self.conn.lookupByName(self.name)
self.dom = dom
+ @staticmethod
+ def dom_details (dom):
+ output=""
+ output += " id=%s - OSType=%s"%(dom.ID(),dom.OSType())
+ # calling state() seems to be working fine
+ (state,reason)=dom.state()
+ output += " state=%s, reason=%s"%(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: maxmem = %s, mem = %s, ncpu = %s, cputime = %s]" % (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 %s"%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"
+ xml = open("/etc/libvirt/lxc/%s.xml" % self.name).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", "/etc/libvirt/lxc/%s.xml" % self.name]
+ logger.log_call(command)
+
def start(self, delay=0):
- ''' Just start the sliver '''
+ '''Just start the sliver'''
logger.verbose('sliver_libvirt: %s start'%(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 %s" % 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))
@@ -95,30 +152,14 @@ class Sliver_Libvirt(Account):
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: %s => %s'%(self,result))
+ return result
def configure(self, rec):
@@ -171,14 +212,19 @@ class Sliver_Libvirt(Account):
# Call the upper configure method (ssh keys...)
Account.configure(self, rec)
+ @staticmethod
+ def get_unique_vif():
+ return 'veth%s' % random.getrandbits(32)
+
# A placeholder until we get true VirtualInterface objects
@staticmethod
def get_interfaces_xml(rec):
xml = """
+
-"""
+""" % (Sliver_Libvirt.get_unique_vif())
try:
tags = rec['rspec']['tags']
if 'interface' in tags:
@@ -198,14 +244,17 @@ class Sliver_Libvirt(Account):
%s
+
- """ % (interface['bridge'], vlanxml)
+ """ % (interface['bridge'], vlanxml, Sliver_Libvirt.get_unique_vif())
else:
tag_xml = tag_xml + """
+
- """
+ """ % (Sliver_Libvirt.get_unique_vif())
+
xml = tag_xml
logger.log('sliver_libvirty.py: interface XML is: %s' % xml)