an attempt at robustifying slice teardown once again

[nodemanager.git] / sliver_lxc.py

diff --git a/sliver_lxc.py b/sliver_lxc.py
index a4f65f0..a08b0ce 100644 (file)
--- a/sliver_lxc.py
+++ b/sliver_lxc.py
@@ -10,6 +10,13 @@ import grp
 from pwd import getpwnam
 from string import Template
 
+# vsys probably should not be a plugin
+# the thing is, the right way to handle stuff would be that
+# if slivers get created by doing a,b,c
+# then they sohuld be delted by doing c,b,a
+# the current ordering model for vsys plugins completely fails to capture that
+from plugins.vsys import removeSliverFromVsys
+
 import libvirt
 
 import logger
@@ -83,7 +90,7 @@ class Sliver_LXC(Sliver_Libvirt, Initscript):
 
         vref = rec['vref']
         if vref is None:
-            vref = "lxc-f14-x86_64"
+            vref = "lxc-f18-x86_64"
             logger.log("sliver_libvirt: %s: WARNING - no vref attached, using hard-wired default %s" % (name,vref))
 
         refImgDir    = os.path.join(Sliver_LXC.REF_IMG_BASE_DIR, vref)
@@ -229,6 +236,10 @@ unset pathmunge
             if not found:
                 with open(from_root,"a") as user_profile:
                     user_profile.write(enforced_line)
+                # in case we create the slice's .profile when writing
+                if from_root.find("/home")>=0:
+                    command=['chown','%s:slices'%name,from_root]
+                    logger.log_call(command,timeout=5)
 
         # Lookup for xid and create template after the user is created so we
         # can get the correct xid based on the name of the slice
@@ -259,7 +270,7 @@ unset pathmunge
             dom = conn.lookupByName(name)
         except:
             dom = conn.defineXML(xml)
-        logger.verbose('lxc_create: %s -> %s'%(name, Sliver_Libvirt.debuginfo(dom)))
+        logger.verbose('lxc_create: %s -> %s'%(name, Sliver_Libvirt.dom_details(dom)))
 
 
     @staticmethod
@@ -271,41 +282,51 @@ unset pathmunge
 
         containerDir = Sliver_LXC.CON_BASE_DIR + '/%s'%(name)
 
+        # Slivers with vsys running will fail the subvolume delete
+        # A more permanent solution may be to ensure that the vsys module
+        # is called before the sliver is destroyed.
+        removeSliverFromVsys (name)
+
         try:
             # Destroy libvirt domain
             dom = conn.lookupByName(name)
         except:
-            logger.verbose('sliver_lxc: Domain %s does not exist!' % name)
+            logger.verbose('sliver_lxc.destroy: Domain %s does not exist!' % name)
+            return
 
         try:
+            logger.log("sliver_lxc.destroy: destroying domain %s"%name)
             dom.destroy()
         except:
-            logger.verbose('sliver_lxc: Domain %s not running... continuing.' % name)
+            logger.verbose('sliver_lxc.destroy: Domain %s not running... continuing.' % name)
 
         try:
+            logger.log("sliver_lxc.destroy: undefining domain %s"%name)
             dom.undefine()
         except:
-            logger.verbose('sliver_lxc: Domain %s is not defined... continuing.' % name)
+            logger.verbose('sliver_lxc.destroy: Domain %s is not defined... continuing.' % name)
 
         # Remove user after destroy domain to force logout
         command = ['/usr/sbin/userdel', '-f', '-r', name]
         logger.log_call(command, timeout=15*60)
 
-        if os.path.exists(os.path.join(containerDir,"vsys")):
-            # Slivers with vsys running will fail the subvolume delete.
-            # A more permanent solution may be to ensure that the vsys module
-            # is called before the sliver is destroyed.
-            logger.log("destroying vsys directory and restarting vsys")
-            logger.log_call(["rm", "-fR", os.path.join(containerDir, "vsys")])
-            logger.log_call(["/etc/init.d/vsys", "restart", ])
-
         # Remove rootfs of destroyed domain
         command = ['btrfs', 'subvolume', 'delete', containerDir]
         logger.log_call(command, timeout=60)
 
-        if os.path.exists(containerDir):
-           # oh no, it's still here...
-           logger.log("WARNING: failed to destroy container %s" % containerDir)
+        if not os.path.exists(containerDir):
+            logger.log('sliver_lxc.destroy: %s cleanly destroyed.'%name)
+        else:
+            # oh no, it's still here...
+            logger.log("sliver_lxc.destroy: 1st warning: could not delete %s" % containerDir)
+            # this is for debugging but does not seem to be of much use
+            logger.log_call (['lsof'])
+            # what I can see on running nodes is that a second subvolume delete seems to do the trick here
+            # so let's check if that could be a workaround
+            logger.log("sliver_lxc.destroy: 2nd attempt at btrfs subvolume delete %s" % containerDir)
+            command = ['btrfs', 'subvolume', 'delete', containerDir]
+            logger.log_call(command, timeout=60)
+            if not os.path.exists(containerDir):
+                logger.log("sliver_lxc.destroy: WARNING: failed to delete %s after 2 attempts"%containerDir)
 
-        logger.verbose('sliver_libvirt: %s destroyed.'%name)