X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=sliver_libvirt.py;h=5407cdc41249ab2aeda55e67ec4f427d9cebee46;hb=ecee05390277f57b02d21ffca0195292bde1defa;hp=98525c08e2f0b6aabb2ec721fe92f7ade6fd0839;hpb=db1f4974bb5e2f250152890207c9164f8fa4a852;p=nodemanager.git

diff --git a/sliver_libvirt.py b/sliver_libvirt.py
index 98525c0..5407cdc 100644
--- a/sliver_libvirt.py
+++ b/sliver_libvirt.py
@@ -2,6 +2,7 @@
 
 import sys
 import os, os.path
+import re
 import subprocess
 import pprint
 import random
@@ -23,6 +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
@@ -39,12 +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))
@@ -69,6 +71,48 @@ 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("<target dev='veth[0-9]*'/>", 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 '''
         logger.verbose('sliver_libvirt: %s start'%(self.name))
@@ -76,7 +120,18 @@ class Sliver_Libvirt(Account):
         # Check if it's running to avoid throwing an exception if the
         # domain was already running, create actually means start
         if not self.is_running():
-            self.dom.create()
+            try:
+                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))
 
@@ -98,33 +153,12 @@ class Sliver_Libvirt(Account):
         except:
             logger.log_exc("in sliver_libvirt.stop",name=self.name)
 
-    def is_running (self):
-        result=self._is_running()
-        logger.log("sliver_libvirt.is_running on %s returned %s"%(self.name,result))
-        return result
-
-    def _is_running(self):
+    def is_running(self):
         ''' Return True if the domain is running '''
-        logger.verbose('sliver_libvirt: entering is_running on [%s:%s]'%(self.name,self.dom.ID()))
-        try:
-            state, _, _, _, _ = self.dom.info()
-            if state == libvirt.VIR_DOMAIN_RUNNING:
-                logger.verbose('sliver_libvirt: %s is RUNNING'%self.name)
-                return True
-            else:
-                info = Sliver_Libvirt.debuginfo(self.dom)
-                logger.verbose('sliver_libvirt: %s is ' \
-                               'NOT RUNNING...\n%s'%(self.name, info))
-                return False
-        except:
-            logger.log("Re-fetching dom from name=%s"%self.name)
-            try:
-                self.dom=self.conn.lookupByName(self.name)
-                state, _, _, _, _ = self.dom.info()
-                return state==libvirt.VIR_DOMAIN_RUNNING
-            except:
-                logger.log_exc("in sliver_libvirt.is_running",name=self.name)
-                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):