X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=sliver_lxc.py;h=7f548b5c196c7de740865b8f11bb704376fc282e;hb=385b6b2088b9e9413e3b85f46adb18d4a38eebbb;hp=37055986dd34ad82073f6e2542b46dd2bddbb86f;hpb=ceae8bc56f3f46d5913ca8988b701337a35135cf;p=nodemanager.git diff --git a/sliver_lxc.py b/sliver_lxc.py index 3705598..7f548b5 100644 --- 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, startService as vsysStartService + 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) @@ -95,6 +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] @@ -263,7 +293,7 @@ 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 @@ -279,37 +309,56 @@ unset pathmunge # 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", ]) - # 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=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: + 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()