Skip to content
Snippets Groups Projects
Commit 3bbd843b authored by David Johnson's avatar David Johnson
Browse files

Several things in this commit: 

  * Prior to this commit, libplab depended on db state to create slivers
    and slices.  Now it can be done using the regular command line tools
    without the metadata in the db.  This makes development and debugging
    much easier and allows us to use the command-line tools even if state has
    been cleared out of the db (i.e., for sliver garbage collection).
  * Add support for sliver start/stop/restart via the v4 NM.
  * Some support for sliver garbage collection.
  * Various other improvements and cleanup.
parent 4a50330d
No related branches found
No related tags found
No related merge requests found
Hide whitespace changes
Inline Side-by-side
tbsetup/plab/libplab.py.in
+ 506
− 197
View file @ 3bbd843b
......@@ -312,6 +312,15 @@ class Plab:
slice._create()
return slice
def createSliceNoDB(self,name,description,userlist=[],nodelist=[]):
"""
Slice factory function that doesn't use the Emulab db.
"""
slice = Slice(self,usedb=False,slicename=name,slicedescr=description,
userlist=userlist,nodelist=nodelist)
slice._create()
return slice
def loadSlice(self, pid, eid):
"""
Slice factory function
......@@ -320,6 +329,14 @@ class Plab:
slice._load()
return slice
def loadSliceNoDB(self,name):
"""
Slice factory function that doesn't use the Emulab db.
"""
slice = Slice(self,usedb=False,slicename=name)
slice._load()
return slice
def updateNodeEntries(self, ignorenew = False):
"""
Finds out which Plab nodes are available, and
......@@ -1149,14 +1166,39 @@ wrap_around(Plab.createSlice, timeAdvice)
#
class Slice:
def __init__(self, plab, pid, eid, slicename = None):
def __init__(self, plab, pid = None, eid = None, exptidx = None,
usedb = True, slicename = None, slicedescr = DEF_SLICE_DESC,
sliceurl = None, userlist = [], nodelist = []):
self.plab = plab
self.pid, self.eid = pid, eid
self.slicemeta = None
self.slicemeta_legacy = None
self.exptidx = None
self.exptidx = exptidx
self.slicename = slicename
self.description = DEF_SLICE_DESC
self.description = slicedescr
self.sliceurl = sliceurl
self.usedb = usedb
self.userlist = userlist
self.nodelist = nodelist
# Rules for arguments.
if usedb and (pid == None or eid == None):
raise RuntimeError, "Must provide pid/eid if using DB!"
if not usedb and slicename == None:
raise RuntimeError, "Must provide slicename if not using DB!"
# For the case where usedb is false, we have to set eid/pid to
# empty strings instead of None, since various other classes assume
# they are set. Same with leaseend -- for that, we just pick the
# current time (it's a safe, if impractical, base for "renewing" the
# slice). We do attempt to retrieve leaseend from PLC in _load.
if self.pid == None:
self.pid = ""
if self.eid == None:
self.eid = ""
self.leaseend = time.gmtime(time.time())
return
def _create(self):
......@@ -1172,23 +1214,25 @@ class Slice:
adminbit = 1
pass
res = DBQueryFatal("select idx, expt_name from experiments "
"where pid=%s "
"and eid=%s",
(self.pid, self.eid))
if not len(res):
raise RuntimeError, \
"Didn't get any results while looking up info on " \
"experiment %s/%s" % (self.pid, self.eid)
(eindex, descr) = res[0]
if not self.slicename:
self.slicename = "%s_%s" % (SLICEPREFIX, eindex)
if self.usedb:
res = DBQueryFatal("select idx, expt_name from experiments "
"where pid=%s "
"and eid=%s",
(self.pid, self.eid))
if not len(res):
raise RuntimeError, \
"Didn't get any results while looking up info on " \
"experiment %s/%s" % (self.pid, self.eid)
(eindex, descr) = res[0]
if not self.slicename:
self.slicename = "%s_%s" % (SLICEPREFIX, eindex)
pass
self.description = descr
self.exptidx = eindex
pass
self.description = descr
self.exptidx = eindex
print "Creating Plab slice %s." % self.slicename
try:
......@@ -1205,8 +1249,8 @@ class Slice:
pass
pass
# see if we have two tickets to insert (support the legacy ticket
# or not)
# see if we have two tickets to insert (support the legacy
# ticket or not)
if type(tmpslicemeta) == tuple:
self.slicemeta = tmpslicemeta[0]
self.slicemeta_legacy = tmpslicemeta[1]
......@@ -1232,10 +1276,12 @@ class Slice:
adminbit)
pass
qstr = "insert into plab_slices " + insertFieldsStr + \
" values " + insertValuesStr
DBQueryFatal(qstr,insertValuesTuple)
if self.usedb:
qstr = "insert into plab_slices " + insertFieldsStr + \
" values " + insertValuesStr
DBQueryFatal(qstr,insertValuesTuple)
pass
pass
except:
self.plab.agent.deleteSlice(self)
......@@ -1249,31 +1295,36 @@ class Slice:
pass
pass
DBQueryFatal("delete from plab_slices where slicename=%s",
(self.slicename,))
if self.usedb:
DBQueryFatal("delete from plab_slices where slicename=%s",
(self.slicename,))
pass
raise
# Setup mailing alias for the slice. All mail currently filters
# in via the 'emulabman' alias. A procmail filter there will
# redirect it to the appropriate user.
try:
qres = DBQueryFatal("select u.uid, u.usr_email from users as u "
"left join experiments as e "
"on u.uid_idx = e.swapper_idx "
"where e.pid=%s and e.eid=%s",
(self.pid, self.eid))
if not len(qres):
raise RuntimeError, \
"Didn't get any results while looking up user info " \
"for experiment %s/%s" % (self.pid, self.eid)
(username, usremail) = qres[0]
command = "%s -host %s /bin/echo %s \> %s/%s" % \
(SSH, USERNODE, usremail,
SLICE_ALIAS_DIR, self.slicename)
os.system(command)
except:
print "Could not setup email alias for slice: %s!" % self.slicename
traceback.print_exc()
if self.usedb:
# Setup mailing alias for the slice. All mail currently filters
# in via the 'emulabman' alias. A procmail filter there will
# redirect it to the appropriate user.
try:
qres = DBQueryFatal("select u.uid, u.usr_email from users as u "
"left join experiments as e "
"on u.uid_idx = e.swapper_idx "
"where e.pid=%s and e.eid=%s",
(self.pid, self.eid))
if not len(qres):
raise RuntimeError, \
"Didn't get any results while looking up user info" \
" for experiment %s/%s" % (self.pid, self.eid)
(username, usremail) = qres[0]
command = "%s -host %s /bin/echo %s \> %s/%s" % \
(SSH, USERNODE, usremail,
SLICE_ALIAS_DIR, self.slicename)
os.system(command)
except:
print "Could not setup email alias for slice: %s!" \
% self.slicename
traceback.print_exc()
pass
pass
return res
......@@ -1286,6 +1337,25 @@ class Slice:
XXX This should probably be made lazy, since not all operations
really need it
"""
if not self.usedb:
# Just grab NM tickets from PLC for now
try:
self.slicemeta = self.plab.agent.getSliceMeta(self)
self.slicemeta_legacy = self.plab.agent.getSliceMetaLegacy(self)
pass
except:
raise
# Grab leaseend time
try:
self.leaseend = self.plab.agent.getSliceExpTime(self)
pass
except:
# do nothing -- we can survive without the leaseend time.
pass
return
if verbose:
print "Loading slice for pid/eid %s/%s" % (self.pid, self.eid)
pass
......@@ -1340,15 +1410,17 @@ class Slice:
"\n watch for inconsistent DB state!"
pass
pass
DBQueryFatal("update plab_slices "
" set slicemeta=%s, leaseend=%s, slicemeta_legacy=%s "
" where slicename=%s",
(self.slicemeta,
time.strftime("%Y-%m-%d %H:%M:%S",
time.gmtime(self.leaseend)),
self.slicemeta_legacy,
self.slicename))
if self.usedb:
DBQueryFatal("update plab_slices "
"set slicemeta=%s,leaseend=%s,slicemeta_legacy=%s"
" where slicename=%s",
(self.slicemeta,
time.strftime("%Y-%m-%d %H:%M:%S",
time.gmtime(self.leaseend)),
self.slicemeta_legacy,
self.slicename))
pass
pass
except:
print "slice.renew: Slice renewal failed:"
......@@ -1382,7 +1454,7 @@ class Slice:
pass
# Report any nodes that are near to expiration
if len(reportfailed) > 0:
if self.usedb and len(reportfailed) > 0:
tbstr = ""
for nodeid, leaseend in reportfailed:
tbstr += "Node: %s, Leaseend: %s UTC\n" % \
......@@ -1405,14 +1477,25 @@ class Slice:
slice is destroyed.
"""
print "Destroying Plab slice %s." % self.slicename
res = DBQueryFatal("select node_id from plab_slice_nodes"
" where slicename = %s",
(self.slicename))
nodeidlist = []
if self.usedb:
res = DBQueryFatal("select node_id from plab_slice_nodes"
" where slicename = %s",
(self.slicename))
for (nodeid,) in res:
nodeidlist.append(nodeid)
pass
pass
else:
nodeidlist = self.nodelist
pass
print "\tRemoving any remaining nodes in slice.."
for (nodeid,) in res:
node = self.loadNode(nodeid)
for nid in nodeidlist:
node = self.loadNode(nid)
node.free()
del node # Encourage the GC'er
pass
osigs = disable_sigs(TERMSIGS)
......@@ -1432,29 +1515,30 @@ class Slice:
"\n watch for inconsistent DB state!"
pass
pass
try:
print "\tRemoving slice DB entry."
DBQueryFatal("delete from plab_slices where slicename = %s",
(self.slicename,))
except:
print "Error deleting slice from DB!"
tbstr = "".join(traceback.format_exception(*sys.exc_info()))
SENDMAIL(TBOPS, "Error deleting slice from DB",
"Slice deletion error:\n\n%s" % tbstr, TBOPS)
enable_sigs(osigs)
raise
try:
command = "%s -host %s /bin/rm -f %s/%s" % \
(SSH, USERNODE, SLICE_ALIAS_DIR, self.slicename)
os.system(command)
except:
print "Could not remove email alias for slice: %s!" % \
self.slicename
traceback.print_exc()
pass
if self.usedb:
try:
print "\tRemoving slice DB entry."
DBQueryFatal("delete from plab_slices where slicename = %s",
(self.slicename,))
except:
print "Error deleting slice from DB!"
tbstr = "".join(traceback.format_exception(*sys.exc_info()))
SENDMAIL(TBOPS, "Error deleting slice from DB",
"Slice deletion error:\n\n%s" % tbstr, TBOPS)
enable_sigs(osigs)
raise
try:
command = "%s -host %s /bin/rm -f %s/%s" % \
(SSH, USERNODE, SLICE_ALIAS_DIR, self.slicename)
os.system(command)
except:
print "Could not remove email alias for slice: %s!" % \
self.slicename
traceback.print_exc()
pass
pass
enable_sigs(osigs)
......@@ -1463,14 +1547,14 @@ class Slice:
Node factory function
"""
# XXX: KRW - The following is a hack to help me with testing.
if not nodeid.startswith("plab"):
IP = socket.gethostbyname(nodeid)
qres = DBQueryFatal("select n.node_id from nodes as n left join "
"interfaces as i on n.node_id = i.node_id "
"where i.IP = %s", (IP,))
assert (len(qres) > 0), "Node does not exist in DB: %s" % nodeid
nodeid = qres[0][0] + "-20"
node = Node(self, nodeid)
#if not nodeid.startswith("plab"):
# IP = socket.gethostbyname(nodeid)
# qres = DBQueryFatal("select n.node_id from nodes as n left join "
# "interfaces as i on n.node_id = i.node_id "
# "where i.IP = %s", (IP,))
# assert (len(qres) > 0), "Node does not exist in DB: %s" % nodeid
# nodeid = qres[0][0] + "-20"
node = Node(self, nodeid, usedb = self.usedb)
node._create(force)
return node
......@@ -1478,7 +1562,7 @@ class Slice:
"""
Node factory function
"""
node = Node(self, nodeid)
node = Node(self, nodeid, usedb = self.usedb)
node._load()
return node
......@@ -1490,9 +1574,13 @@ class Slice:
# Handle legacy tickets for compat purposes.
self.slicemeta = self.plab.agent.getSliceMeta(self)
self.slicemeta_legacy = self.plab.agent.getSliceMetaLegacy(self)
DBQueryFatal("update plab_slices set "
"slicemeta=%s,slicemeta_legacy=%s where slicename=%s",
(self.slicemeta,self.slicemeta_legacy,self.slicename))
if self.usedb:
DBQueryFatal("update plab_slices set"
" slicemeta=%s,slicemeta_legacy=%s"
" where slicename=%s",
(self.slicemeta,self.slicemeta_legacy,
self.slicename))
pass
pass
except:
print "Error updating slice metadata!"
......@@ -1505,33 +1593,83 @@ class Slice:
"""
Return a tuple containing the nodes that belong to this slice
"""
# Grab set of plab nodes belonging to expt and their IPs:
return DBQueryFatal("select r.node_id, i.IP, w.hostname "
" from reserved as r "
" left join nodes as n1 "
" on r.node_id = n1.node_id "
" left join nodes as n2 "
" on n1.phys_nodeid = n2.node_id "
" left join widearea_nodeinfo as w "
" on n2.node_id = w.node_id "
" left join interfaces as i "
" on w.node_id = i.node_id "
" where r.pid=%s and r.eid=%s "
" and n1.type=%s and i.role=%s",
(self.pid, self.eid, "pcplab", "ctrl"))
if self.usedb:
# Grab set of plab nodes belonging to expt and their IPs:
return DBQueryFatal("select r.node_id, i.IP, w.hostname "
" from reserved as r "
" left join nodes as n1 "
" on r.node_id = n1.node_id "
" left join nodes as n2 "
" on n1.phys_nodeid = n2.node_id "
" left join widearea_nodeinfo as w "
" on n2.node_id = w.node_id "
" left join interfaces as i "
" on w.node_id = i.node_id "
" where r.pid=%s and r.eid=%s "
" and n1.type=%s and i.role=%s",
(self.pid, self.eid, "pcplab", "ctrl"))
else:
retval = []
for n in self.nodelist:
nn = Node(self,n)
# Grab the IP and hostname
retval.append([n,nn.IP,nn.hostname])
pass
return retval
pass
def getSliceUsers(self):
return self.userlist
pass # end of class Slice
# AOP wrappers for class Slice
wrap_around(Slice._create, timeAdvice)
wrap_around(Slice.destroy, timeAdvice)
# Helper regexps used in the Node class
VNODE_REG = re.compile("^plabvm(\d+)\-(\d+)$")
PNODE_REG = re.compile("^plab(\d+)$")
IP_REG = re.compile("^(\d+)\.(\d+)\.(\d+)\.(\d+)$")
HOSTNAME_REG = re.compile("^([\d\w\-\.]+)$")
#
# Node abstraction
#
class Node:
def __init__(self, slice, nodeid, pollNode = False):
#
# Note: you can now pass in an Emulab vnode or pnode, or a hostname or
# IP address.
# Also, note that passing in a pnode or hostname/IP forces usedb = False,
# since most Emulab db entries depend on the vnodeid at this level.
#
def __init__(self, slice, nodeid, pollNode = False, usedb = True):
self.usedb = usedb
self.slice = slice
self.nodeid = nodeid
# Figure out what kind of nodeid we were given (how we do lookups
# depends on it):
if VNODE_REG.match(nodeid) != None:
self.nidtype = 'v'
pass
elif PNODE_REG.match(nodeid) != None:
self.nidtype = 'p'
self.usedb = False
pass
elif IP_REG.match(nodeid) != None:
self.nidtype = 'i'
self.usedb = False
pass
elif HOSTNAME_REG.match(nodeid) != None:
self.nidtype = 'h'
self.usedb = False
pass
else:
raise RuntimeError, "nodeid must be an Emulab vnode or pnode, or" \
" a hostname or IP address!"
(self.IP,self.hostname,self.phys_nodeid) = self.__findHostInfo()
self.leaseend = 0
self.nodemeta = None
self.pollNode = pollNode
......@@ -1554,6 +1692,27 @@ class Node:
(component,operation,status,msg)
pass
pass
def __gcNode(self,add_remove):
if add_remove:
# add
now = time.strftime("%Y-%m-%d %H:%M:%S",
time.localtime(time.time()))
q = "replace into plab_sliver_garbage (pid,eid,exptidx," \
" slicename,node_id,phys_node_id,ctime,mtime)"\
" values (%s,%s,%s,%s,%s,%s,%s,%s)"
DBQueryWarn(q,(self.slice.pid,self.slice.eid,self.slice.exptidx,
self.slice.slicename,self.nodeid,self.phys_nodeid,
now,now))
else:
# remove, and don't worry about failed queries, since the garbage
# collector will check for this case anyway.
q = "delete from plab_sliver_garbage" \
" where pid=%s and eid=%s and slicename=%s and node_id=%s"
DBQuery(q,(self.slice.pid,self.slice.eid,self.slice.slicename,
self.phys_nodeid))
pass
# XXX: may want to rethink signal handling here.
def _create(self, force=False):
......@@ -1563,18 +1722,29 @@ class Node:
node setup is performed. Don't call this directly, use
Slice.createNode instead.
"""
# First, make sure there isn't already an entry in the DB
try:
self._load()
except:
pass
else:
if force:
print "Node entry exists in DB, but creation forced anyway."
if self.usedb:
# First, make sure there isn't already an entry in the DB
try:
self._load()
except:
pass
else:
raise RuntimeError, "Entry for plab node %s already exists " \
"in the DB" % self.nodeid
if force:
print "Node entry exists in DB, but creation forced anyway."
else:
raise RuntimeError, "Entry for plab node %s already " \
"exists in the DB" % self.nodeid
pass
pass
if self.usedb:
# Note: whenever we create a plab sliver, we remove the node from
# the plab_sliver_garbage table before even trying the create.
# The reason for this is that even if the create never succeeds
# in a swapin, there will be a corresponding free call for all
# slivers that matter.
self.__gcNode(False)
pass
print "Creating Plab node %s on %s." % (self.nodeid, self.IP)
......@@ -1582,21 +1752,28 @@ class Node:
try:
res, self.nodemeta, self.leaseend = \
self.slice.plab.agent.createNode(self)
self.__logNodeHist('node','create','success','')
if self.usedb:
self.__logNodeHist('node','create','success','')
pass
except:
self.__logNodeHist('node','create','failure',
traceback.format_exception(*sys.exc_info()))
if self.usedb:
self.__logNodeHist('node','create','failure',
traceback.format_exception(*sys.exc_info()))
pass
raise
if self.usedb:
DBQueryFatal("replace into plab_slice_nodes"
" (exptidx, pid, eid, slicename, node_id,"
" nodemeta, leaseend)"
" values (%s, %s, %s, %s, %s, %s, %s)",
(self.slice.exptidx, self.slice.pid, self.slice.eid,
self.slice.slicename, self.nodeid,
self.nodemeta,
time.strftime("%Y-%m-%d %H:%M:%S",
time.gmtime(self.leaseend))))
pass
DBQueryFatal("replace into plab_slice_nodes"
" (exptidx, pid, eid, slicename, node_id,"
" nodemeta, leaseend)"
" values (%s, %s, %s, %s, %s, %s, %s)",
(self.slice.exptidx, self.slice.pid, self.slice.eid,
self.slice.slicename, self.nodeid,
self.nodemeta,
time.strftime("%Y-%m-%d %H:%M:%S",
time.gmtime(self.leaseend))))
if self.pollNode:
TIMESTAMP("Waiting for %s to respond" % self.nodeid)
while True:
......@@ -1613,12 +1790,15 @@ class Node:
TIMESTAMP("createnode finished on %s." % self.nodeid)
return
def _load(self):
"""
Loads an already allocated node from the DB. Don't call this
directly, use Slice.loadNode instead.
"""
if not self.usedb:
return
if verbose:
print "Loading node %s" % self.nodeid
res = DBQueryFatal("select slicename, nodemeta, leaseend "
......@@ -1663,27 +1843,36 @@ class Node:
deleted = 0
TIMESTAMP("freenode %s started." % self.nodeid)
print "Freeing Plab node %s." % self.nodeid
# Remove the DB entry first.
try:
DBQueryFatal("delete from plab_slice_nodes where node_id = %s",
(self.nodeid,))
except:
print "Uh oh, couldn't remove plab sliver record from the DB!"
tbstr = "".join(traceback.format_exception(*sys.exc_info()))
SENDMAIL(TBOPS, "Error: Couldn't remove plab vnode from DB",
"Unable to delete entry for sliver %s from the DB:"
"\n\n%s" % (self.nodeid, tbstr), TBOPS)
if self.usedb:
# Remove the DB entry first.
try:
DBQueryFatal("delete from plab_slice_nodes where node_id = %s",
(self.nodeid,))
except:
print "Uh oh, couldn't remove plab sliver record from the DB!"
tbstr = "".join(traceback.format_exception(*sys.exc_info()))
SENDMAIL(TBOPS, "Error: Couldn't remove plab vnode from DB",
"Unable to delete entry for sliver %s from the DB:"
"\n\n%s" % (self.nodeid, tbstr), TBOPS)
pass
pass
deleted = 0
try:
deleted = self.slice.plab.agent.freeNode(self)
# Uncomment to increase logging
#self.__logNodeHist('node','free','success','')
if self.usedb:
# Uncomment to increase logging
#self.__logNodeHist('node','free','success','')
pass
except:
self.__logNodeHist('node','free','failure',
traceback.format_exception(*sys.exc_info()))
# since the free failed, we need to add this
# node to plab_sliver_garbage
if self.usedb:
self.__gcNode(True)
self.__logNodeHist('node','free','failure',
traceback.format_exception(*sys.exc_info()))
pass
raise
TIMESTAMP("freenode %s finished." % self.nodeid)
......@@ -1704,22 +1893,110 @@ class Node:
try:
res, self.nodemeta, self.leaseend = \
self.slice.plab.agent.renewNode(self)
self.__logNodeHist('node','renew','success','')
if self.usedb:
self.__logNodeHist('node','renew','success','')
pass
except:
self.__logNodeHist('node','renew','failure',
traceback.format_exception(*sys.exc_info()))
if self.usedb:
self.__logNodeHist('node','renew','failure',
traceback.format_exception(*sys.exc_info()))
pass
raise
if self.usedb:
DBQueryFatal("update plab_slice_nodes"
" set nodemeta = %s, leaseend = %s"
" where node_id = %s",
(self.nodemeta,
time.strftime("%Y-%m-%d %H:%M:%S",
time.gmtime(self.leaseend)),
self.nodeid))
pass
DBQueryFatal("update plab_slice_nodes"
" set nodemeta = %s, leaseend = %s"
" where node_id = %s",
(self.nodemeta,
time.strftime("%Y-%m-%d %H:%M:%S",
time.gmtime(self.leaseend)),
self.nodeid))
TIMESTAMP("renewnode %s finished." % self.nodeid)
return 0
def start(self):
"""
Start up this node via the NM Start call (start means run the main
vserver init script). Note that this method forks and runs another
private method to actually do the work!
"""
res = ForkCmd(self._start, timeout = RENEW_TIMEOUT,
disable_sigs_parent = TERMSIGS)
return res[0] | res[1]
def _start(self):
res = None
try:
res = self.slice.plab.agent.startNode(self)
if self.usedb:
self.__logNodeHist('node','start','success','')
pass
except:
if self.usedb:
self.__logNodeHist('node','start','failure',
traceback.format_exception(*sys.exc_info()))
pass
raise
TIMESTAMP("startnode %s finished." % self.nodeid)
return 0
def stop(self):
"""
Stop this node via the NM Start call (start means kill all processes
in the vserver). Note that this method forks and runs another
private method to actually do the work!
"""
res = ForkCmd(self._stop, timeout = RENEW_TIMEOUT,
disable_sigs_parent = TERMSIGS)
return res[0] | res[1]
def _stop(self):
res = None
try:
res = self.slice.plab.agent.stopNode(self)
if self.usedb:
self.__logNodeHist('node','stop','success','')
pass
except:
if self.usedb:
self.__logNodeHist('node','stop','failure',
traceback.format_exception(*sys.exc_info()))
pass
raise
TIMESTAMP("stopnode %s finished." % self.nodeid)
return 0
def restart(self):
"""
Restarts this node via the NM Start call (restart means kill all
processes in the vserver and run /etc/rc.vinit). Note that this
method forks and runs another private method to actually do the work!
"""
res = ForkCmd(self._restart, timeout = RENEW_TIMEOUT,
disable_sigs_parent = TERMSIGS)
return res[0] | res[1]
def _restart(self):
res = None
try:
res = self.slice.plab.agent.restartNode(self)
if self.usedb:
self.__logNodeHist('node','restart','success','')
pass
except:
if self.usedb:
self.__logNodeHist('node','restart','failure',
traceback.format_exception(*sys.exc_info()))
pass
raise
TIMESTAMP("restartnode %s finished." % self.nodeid)
return 0
def emulabify(self, rootballpath = DEFAULT_DATA_PATH,
rootballname = DEF_ROOTBALL_NAME):
"""
......@@ -1782,11 +2059,15 @@ class Node:
pass
try:
self.unpackRootball(rootballpath, rrootballname)
# Uncomment to increase logging
#self.__logNodeHist('node','emulabify','success','')
if self.usedb:
# Uncomment to increase logging
#self.__logNodeHist('node','emulabify','success','')
pass
except:
self.__logNodeHist('node','emulabify','failure',
traceback.format_exception(*sys.exc_info()))
if self.usedb:
self.__logNodeHist('node','emulabify','failure',
traceback.format_exception(*sys.exc_info()))
pass
raise
TIMESTAMP("emulabify finished on %s." % self.nodeid)
......@@ -1947,42 +2228,57 @@ class Node:
"""
Figures out and returns the IP of the remote node.
"""
res = DBQueryFatal("select i.IP from nodes as nv"
" left join interfaces as i on"
" nv.phys_nodeid=i.node_id"
" where nv.node_id=%s"
" limit 1",
(self.nodeid))
if (not res or len(res) == 0):
# XXX: send email
print "Warning: no IP found for nodeid %s" % self.nodeid
IP = "0.0.0.0"
if self.nidtype == 'v' or self.nidtype == 'p':
res = DBQueryFatal("select i.IP from nodes as nv"
" left join interfaces as i on"
" nv.phys_nodeid=i.node_id"
" where nv.node_id=%s"
" limit 1",
(self.nodeid))
if (not res or len(res) == 0):
# XXX: send email
print "Warning: no IP found for nodeid %s" % self.nodeid
IP = "0.0.0.0"
pass
else:
((IP, ), ) = res
pass
pass
else:
((IP, ), ) = res
IP = socket.gethostbyname(self.nodeid)
pass
if debug:
print "IP is %s for node %s" % (IP, self.nodeid)
return IP
def __findHostname(self):
"""
Grabs the publicly-routable hostname of the remote node.
"""
res = DBQueryFatal("select pm.hostname,i.IP from nodes as nv"
" left join interfaces as i"
" on nv.phys_nodeid=i.node_id"
" left join plab_mapping as pm"
" on i.IP=pm.IP"
" where nv.node_id='%s'" % (self.nodeid))
if (not res or len(res) == 0):
print "Warning: no hostname found for nodeid %s" % self.nodeid
hostname = None
if self.nidtype == 'v' or self.nidtype == 'p':
res = DBQueryFatal("select pm.hostname,i.IP from nodes as nv"
" left join interfaces as i"
" on nv.phys_nodeid=i.node_id"
" left join plab_mapping as pm"
" on i.IP=pm.IP"
" where nv.node_id='%s'" % (self.nodeid))
if (not res or len(res) == 0):
print "Warning: no hostname found for nodeid %s" % self.nodeid
hostname = None
pass
else:
((hostname,IP),) = res
pass
pass
elif self.nidtype == 'i':
hostname = socket.gethostbyaddr(self.nodeid)
pass
else:
((hostname,IP),) = res
hostname = self.nodeid
pass
if debug:
print "hostname is %s for node %s" % (hostname,IP)
pass
......@@ -1997,20 +2293,33 @@ class Node:
Grabs the publicly-routable IP and hostname of the remote node,
and also our phys_nodeid for it.
"""
res = DBQueryFatal("select i.IP,pm.hostname,nv.phys_nodeid "
" from nodes as nv"
" left join interfaces as i"
" on nv.phys_nodeid=i.node_id"
" left join plab_mapping as pm"
" on i.IP=pm.IP"
" where nv.node_id='%s'" % (self.nodeid))
if (not res or len(res) == 0):
print "Warning: no hostinfo found for nodeid %s" % self.nodeid
(IP,hostname,phys_nodeid) = (None,None,None)
if self.nidtype == 'v' or self.nidtype == 'p':
res = DBQueryFatal("select i.IP,pm.hostname,nv.phys_nodeid "
" from nodes as nv"
" left join interfaces as i"
" on nv.phys_nodeid=i.node_id"
" left join plab_mapping as pm"
" on i.IP=pm.IP"
" where nv.node_id='%s'" % (self.nodeid))
if (not res or len(res) == 0):
print "Warning: no hostinfo found for nodeid %s" % self.nodeid
(IP,hostname,phys_nodeid) = (None,None,None)
pass
else:
((IP,hostname,phys_nodeid),) = res
pass
pass
elif self.nidtype == 'i':
IP = self.nodeid
hostname = socket.gethostbyaddr(IP)
phys_nodeid = None
pass
else:
((IP,hostname,phys_nodeid),) = res
hostname = self.nodeid
IP = socket.gethostbyname(self.nodeid)
phys_nodeid = None
pass
if debug:
print "hostname is %s for node %s" % (hostname,IP)
pass
......
tbsetup/plab/mod_PLC4.py.in
+ 60
− 15
View file @ 3bbd843b
......@@ -306,9 +306,9 @@ class NMagent_wrapper:
arg = ticketdata
pass
# if NM4agent, try delivering the ticket first, to maximize our
# chances:
if self.__agent.__class__ == NM4agent:
# If the agent wants to deliver a ticket explicitly before the create
# call, we do that.
if self.__agent.__class__.__dict__.has_key('deliver_ticket'):
try:
res = tryXmlrpcCmd(self.__agent.deliver_ticket,ticketdata)
if res[0] == 0:
......@@ -355,6 +355,21 @@ class NMagent_wrapper:
else:
return tryXmlrpcCmd(self.__agent.create_sliver,ticketdata)
return None
def stop_sliver(self,slicename):
if self.__agent.__class__.__dict__.has_key('stop_sliver'):
return tryXmlrpcCmd(self.__agent.stop_sliver,slicename)
return 0
def start_sliver(self,slicename):
if self.__agent.__class__.__dict__.has_key('start_sliver'):
return tryXmlrpcCmd(self.__agent.start_sliver,slicename)
return 0
def restart_sliver(self,slicename):
if self.__agent.__class__.__dict__.has_key('restart_sliver'):
return tryXmlrpcCmd(self.__agent.restart_sliver,slicename)
return 0
pass
......@@ -421,7 +436,6 @@ class PLCagent:
return self.__server.UpdateSlice(self.__auth, self.__slicename,
{ 'expires' : expdate })
# XXX: this should work ok if xmlrpc converts lists to arrays...
def SliceNodesAdd(self,nodelist):
if not type(nodelist) == list:
nodelist = [nodelist]
......@@ -562,6 +576,7 @@ class mod_PLC4:
now = calendar.timegm(time.gmtime())
try:
# XXX: fix to take desc and url args! (i.e., SliceUpdate)
res = tryXmlrpcCmd(agent.SliceCreate)
if debug:
print "SliceCreate result: %s" % res
......@@ -572,8 +587,16 @@ class mod_PLC4:
raise
try:
userlist = slice.getSliceUsers()
# XXX: we might not always want to add emulabman
if userlist == None:
userlist = []
pass
if not EMULABMAN_EMAIL in userlist:
userlist.append(EMULABMAN_EMAIL)
pass
res = tryXmlrpcCmd(agent.SliceUsersAdd,
EMULABMAN_EMAIL)
userlist)
if debug:
print "SliceUsersAdd result: %s" % res
pass
......@@ -587,9 +610,11 @@ class mod_PLC4:
# that the slice doesn't include the local NM's node... or something
# like that... so we must add the nodes to the slice.
try:
# XXX: test hack
#nodelist = ['pcwf3.emulab.net']
nodelist = map(lambda x: x[2], slice.getSliceNodes())
tnodelist = slice.getSliceNodes()
if tnodelist == None:
tnodelist = []
pass
nodelist = map(lambda x: x[2], tnodelist)
res = tryXmlrpcCmd(agent.SliceNodesAdd, nodelist)
if debug:
print "SliceNodesAdd result: %s" % res
......@@ -756,7 +781,6 @@ class mod_PLC4:
raise
return cPickle.dumps(retval)
# XXX: fix to use new NM
def createNode(self, node):
plcagent = self.__getAgent(node.slice.slicename)
......@@ -820,7 +844,7 @@ class mod_PLC4:
try:
rcap = cPickle.loads(node.nodemeta)
except:
print "WARNING: couldn't load rcap"
#print "WARNING: couldn't load rcap"
pass
node.nmagent = NMagent_wrapper(node.IP,node.nodeid)
res = None
......@@ -840,18 +864,39 @@ class mod_PLC4:
return res
# XXX: MIGHT need fixing...
def renewNode(self, node, length = 0):
return self.createNode(node)
# XXX: add, now that the NM can do this...
def startNode(self,node):
return None
node.nmagent = NMagent_wrapper(node.IP,node.nodeid)
try:
res = node.nmagent.start_sliver(node.slice.slicename)
except:
print "Failed to start node %s/%s from slice %s" % \
(node.nodeid,node.IP,node.slice.slicename)
raise
return res
# XXX: add, now that the NM can do this...
def stopNode(self,node):
return None
node.nmagent = NMagent_wrapper(node.IP,node.nodeid)
try:
res = node.nmagent.stop_sliver(node.slice.slicename)
except:
print "Failed to stop node %s/%s from slice %s" % \
(node.nodeid,node.IP,node.slice.slicename)
raise
return res
def restartNode(self,node):
node.nmagent = NMagent_wrapper(node.IP,node.nodeid)
try:
res = node.nmagent.restart_sliver(node.slice.slicename)
except:
print "Failed to stop node %s/%s from slice %s" % \
(node.nodeid,node.IP,node.slice.slicename)
raise
return res
# XXX: this is broken, appears to be bug in GetSlices xmlrpc call in PLC4
def getSliceExpTime(self, slicename):
"""
......
tbsetup/plab/plabnode.in
+ 70
− 7
View file @ 3bbd843b
......@@ -20,7 +20,10 @@ import libplab
def usage(me):
print "Usage: %s [ -vdf ] { alloc | renew | free } pid eid nodeid" % me
print "Usage: %s [-vdfE] {alloc|renew|free|stop|start|restart} pid eid nodeid\n" \
" Without Emulab DB:\n" \
" %s [-vdfE] {alloc|renew|free|stop|start|restart} slicename nodeid" \
% (me,me)
sys.exit(1)
def processException(command, nodeid, e):
......@@ -46,18 +49,32 @@ def main(args):
parser.add_option("--noIS", dest="noIS", action="store_true",
default=False,
help="Don't run InstantiateSliver() in mod_PLC")
parser.add_option("-E","--no-emulabify",dest="no_emulabify",default=False,
action="store_true",help="Do not install/config Emulab" \
" software on the node")
# parse command line options
opts, args = parser.parse_args(args[1:])
slicename = None
# make sure we were told what to do on the command line.
if not len(args) == 4:
if len(args) == 3:
command, slicename, nodeid = args
pass
elif len(args) == 4:
command, pid, eid, nodeid = args
pass
else:
usage(me)
pass
command, pid, eid, nodeid = args
plab = libplab.Plab()
slice = plab.loadSlice(pid, eid)
if slicename != None:
slice = plab.loadSliceNoDB(slicename)
pass
else:
slice = plab.loadSlice(pid, eid)
pass
if command == "alloc":
try:
......@@ -65,8 +82,10 @@ def main(args):
# With the v4 NM, we have to sleep a couple seconds to give
# the slice the a chance to get keys/acct stuff straightened out
# so we can actually slogin and copy fixsudo.sh to the node.
time.sleep(2)
node.emulabify()
if not opts.no_emulabify:
time.sleep(2)
node.emulabify()
pass
# Note that vnode_setup boots the node
pass
except Exception, e:
......@@ -99,6 +118,50 @@ def main(args):
pass
pass
elif command == "stop":
try:
node = slice.loadNode(nodeid)
node.stop()
pass
except Exception, e:
processException(command, nodeid, e)
sys.exit(1)
pass
pass
elif command == "start":
try:
node = slice.loadNode(nodeid)
node.start()
pass
except Exception, e:
processException(command, nodeid, e)
sys.exit(1)
pass
pass
elif command == "stop":
try:
node = slice.loadNode(nodeid)
node.stop()
pass
except Exception, e:
processException(command, nodeid, e)
sys.exit(1)
pass
pass
elif command == "restart":
try:
node = slice.loadNode(nodeid)
node.restart()
pass
except Exception, e:
processException(command, nodeid, e)
sys.exit(1)
pass
pass
else:
usage(me)
pass
......
tbsetup/plab/plabslice.in
+ 54
− 7
View file @ 3bbd843b
......@@ -12,26 +12,73 @@ import getopt
from libtestbed import *
import libplab
usage = "%prog [ -vd ] { create | destroy } pid eid"
usage = "%prog [-vd] {create|destroy} pid eid\n" \
" Without Emulab DB:\n" \
"%prog [-vd] [-u <user1,...>] [-n <node1,...>] -c <slicedesc> \\\n" \
" {create|destroy} slicename"
# [-w <url>]
def main(args):
me = args[0]
parser = TBParser(usage)
parser.add_option("-u","--users",dest="users",action="store",
help="Comma-separated list of users to add to the slice",
default=None)
parser.add_option("-n","--nodes",dest="nodes",action="store",
help="Comma-separated list of hostnames to add to " \
"the slice",default=None)
parser.add_option("-c","--slicedesc",dest="slicedesc",action="store",
help="Slice description",default=None)
# parser.add_option("-w","--sliceurl",dest="sliceurl",action="store_true",
# help="Slice URL",default=None)
userlist,nodelist = None,None
(opts, args) = parser.parse_args()
if len(args) < 3:
usedb = True
if len(args) == 2:
usedb = False
command,slicename = args
if args[0] == 'create' and (opts.slicedesc == None \
or opts.slicedesc == ''):
parser.error("Must supply a slice description if creating by slicename!")
pass
if opts.nodes != None:
nodelist = opts.nodes.split(',')
pass
if opts.users != None:
userlist = opts.nodes.split(',')
pass
pass
elif len(args) == 3:
command,pid,eid = args
pass
else:
parser.error("Incorrect number of arguments")
command,pid,eid = args
pass
plab = libplab.Plab()
if command == "create":
slice = plab.createSlice(pid, eid)
if usedb:
slice = plab.createSlice(pid, eid)
pass
else:
slice = plab.createSliceNoDB(slicename,opts.slicedesc,
userlist=userlist,nodelist=nodelist)
pass
pass
elif command == "destroy":
slice = plab.loadSlice(pid, eid)
if usedb:
slice = plab.loadSlice(pid, eid)
pass
else:
slice = plab.loadSliceNoDB(slicename)
pass
slice.destroy()
pass
else:
usage(me)
pass
if __name__ == "__main__":
main(sys.argv)
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment