X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=sliver_lxc.py;h=7f548b5c196c7de740865b8f11bb704376fc282e;hb=385b6b2088b9e9413e3b85f46adb18d4a38eebbb;hp=a08b0ce693dd7ab8434aa7136d3e15ab3d585746;hpb=63546278c1e00ac6d1910d4d552a933bd170ed28;p=nodemanager.git diff --git a/sliver_lxc.py b/sliver_lxc.py index a08b0ce..7f548b5 100644 --- a/sliver_lxc.py +++ b/sliver_lxc.py @@ -15,7 +15,7 @@ from string import Template # 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 +from plugins.vsys import removeSliverFromVsys, startService as vsysStartService import libvirt @@ -102,6 +102,29 @@ class Sliver_LXC(Sliver_Libvirt, Initscript): 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/ + # 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 + if not os.path.exists (containerDir): + pass + else: + # if it's empty then let's clean it up + if not os.listdir(containerDir): + # clean up rootfs as userdel will only take care of /home/ + logger.log("sliver_lxc: %s: WARNING cleaning up empty %s"%(name,containerDir)) + command = ['btrfs', 'subvolume', 'delete', containerDir] + logger.log_call(command, timeout=60) + # 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 + else: + logger.log('sliver_lxc: %s: ERROR Could not create sliver - could not clean up pre-existing %s'%(name,containerDir)) + return + # Snapshot the reference image fs (assume the reference image is in its own # subvolume) command = ['btrfs', 'subvolume', 'snapshot', refImgDir, containerDir] @@ -282,11 +305,6 @@ 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) @@ -294,6 +312,10 @@ unset pathmunge 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() @@ -312,21 +334,31 @@ unset pathmunge # Remove rootfs of destroyed domain command = ['btrfs', 'subvolume', 'delete', containerDir] - logger.log_call(command, timeout=60) + logger.log_call(command, timeout=10) + + # 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: - # 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.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()