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, startService as vsysStartService
+
import libvirt
import logger
from account import Account
from sliver_libvirt import Sliver_Libvirt
+BTRFS_TIMEOUT=15*60
+
class Sliver_LXC(Sliver_Libvirt, Initscript):
"""This class wraps LXC commands"""
logger.log('sliver_lxc: %s: ERROR Expected reference image in %s'%(name,refImgDir))
return
+ # in fedora20 we have some difficulty in properly cleaning up /vservers/<slicename>
+ # also note that running e.g. btrfs subvolume create /vservers/.lvref/image /vservers/foo
+ # behaves differently, whether /vservers/foo exists or not:
+ # if /vservers/foo does not exist, it creates /vservers/foo
+ # but if it does exist, then it creates /vservers/foo/image !!
+ # so we need to check the expected container rootfs does not exist yet
+ # this hopefully could be removed in a future release
+ if os.path.exists (containerDir):
+ logger.log("sliver_lxc: %s: WARNING cleaning up pre-existing %s"%(name,containerDir))
+ command = ['btrfs', 'subvolume', 'delete', containerDir]
+ logger.log_call(command, BTRFS_TIMEOUT)
+ # re-check
+ if os.path.exists (containerDir):
+ logger.log('sliver_lxc: %s: ERROR Could not create sliver - could not clean up empty %s'%(name,containerDir))
+ return
+
# Snapshot the reference image fs (assume the reference image is in its own
# subvolume)
command = ['btrfs', 'subvolume', 'snapshot', refImgDir, containerDir]
- if not logger.log_call(command, timeout=15*60):
+ if not logger.log_call(command, timeout=BTRFS_TIMEOUT):
logger.log('sliver_lxc: ERROR Could not create BTRFS snapshot at', containerDir)
return
command = ['chmod', '755', containerDir]
- logger.log_call(command, timeout=15*60)
+ logger.log_call(command)
# TODO: set quotas...
group = grp.getgrnam('slices')
except:
command = ['/usr/sbin/groupadd', 'slices']
- logger.log_call(command, timeout=15*60)
+ logger.log_call(command)
# Add unix account (TYPE is specified in the subclass)
command = ['/usr/sbin/useradd', '-g', 'slices', '-s', Sliver_LXC.SHELL, name, '-p', '*']
- logger.log_call(command, timeout=15*60)
+ logger.log_call(command)
command = ['mkdir', '/home/%s/.ssh'%name]
- logger.log_call(command, timeout=15*60)
+ logger.log_call(command)
# Create PK pair keys to connect from the host to the guest without
# password... maybe remove the need for authentication inside the
# guest?
command = ['su', '-s', '/bin/bash', '-c', 'ssh-keygen -t rsa -N "" -f /home/%s/.ssh/id_rsa'%(name)]
- logger.log_call(command, timeout=60)
+ logger.log_call(command)
command = ['chown', '-R', '%s.slices'%name, '/home/%s/.ssh'%name]
- logger.log_call(command, timeout=30)
+ logger.log_call(command)
command = ['mkdir', '%s/root/.ssh'%containerDir]
- logger.log_call(command, timeout=10)
+ logger.log_call(command)
command = ['cp', '/home/%s/.ssh/id_rsa.pub'%name, '%s/root/.ssh/authorized_keys'%containerDir]
- logger.log_call(command, timeout=30)
+ logger.log_call(command)
logger.log("creating /etc/slicename file in %s" % os.path.join(containerDir,'etc/slicename'))
try:
if uid is not None:
logger.log("uid is %d" % uid)
command = ['mkdir', '%s/home/%s' % (containerDir, name)]
- logger.log_call(command, timeout=10)
+ logger.log_call(command)
command = ['chown', name, '%s/home/%s' % (containerDir, name)]
- logger.log_call(command, timeout=10)
+ logger.log_call(command)
etcpasswd = os.path.join(containerDir, 'etc/passwd')
etcgroup = os.path.join(containerDir, 'etc/group')
if os.path.exists(etcpasswd):
# 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)
+ logger.log_call(command)
# 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
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
# 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
+
+ # Slivers with vsys running will fail the subvolume delete
+ # removeSliverFromVsys return True if it stops vsys, telling us to start it again later
+ vsys_stopped = removeSliverFromVsys (name)
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", ])
+ logger.log_call(command)
# 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)
-
- logger.verbose('sliver_libvirt: %s destroyed.'%name)
-
+ logger.log_call(command, timeout=BTRFS_TIMEOUT)
+
+ # For some reason I am seeing this :
+ #log_call: running command btrfs subvolume delete /vservers/inri_sl1
+ #log_call: ERROR: cannot delete '/vservers/inri_sl1' - Device or resource busy
+ #log_call: Delete subvolume '/vservers/inri_sl1'
+ #log_call:end command (btrfs subvolume delete /vservers/inri_sl1) returned with code 1
+ #
+ # something must have an open handle to a file in there, but I can't find out what it is
+ # the following code aims at gathering data on what is going on in the system at this point in time
+ # note that some time later (typically when the sliver gets re-created) the same
+ # attempt at deleting the subvolume does work
+ # also lsof never shows anything relevant; this is painful..
+
+ if not os.path.exists(containerDir):
+ logger.log('sliver_lxc.destroy: %s cleanly destroyed.'%name)
+ else:
+ logger.log("-TMP-cwd %s : %s"%(name,os.getcwd()))
+ logger.log("-TMP-lsof %s"%name)
+ command=['lsof']
+ logger.log_call(command)
+ logger.log("-TMP-ls-l %s"%name)
+ command = ['ls', '-l', containerDir]
+ logger.log_call(command)
+ if os.path.exists(containerDir):
+ logger.log('sliver_lxc.destroy: ERROR could not cleanly destroy %s - giving up'%name)
+
+ if vsys_stopped: vsysStartService()
git://git.onelab.eu / nodemanager.git / blobdiff