#
# Mark Huang <mlhuang@cs.princeton.edu>
# Andy Bavier <acb@cs.princeton.edu>
+# Faiyaz Ahmed <faiyaza@cs.princeton.edu>
# Copyright (C) 2004-2006 The Trustees of Princeton University
#
-# $Id: bwmon.py,v 1.4 2006/06/02 04:00:00 mlhuang Exp $
+# $Id: bwmon.py,v 1.8 2006/07/10 20:22:12 faiyaza Exp $
#
import syslog
name - slice name
time - beginning of recording period in UNIX seconds
bytes - low bandwidth bytes transmitted at the beginning of the recording period
- exemptbytes - high bandwidth bytes transmitted at the beginning of the recording period
- avgrate - average low bandwidth rate to enforce over the recording period
- avgexemptrate - average high bandwidth rate to enforce over the recording period
+ exemptbytes - high bandwidth bytes transmitted at the beginning of the recording period (for I2 -F)
+ last_avgrate - last recorded avgrate from NM
+ last_maxrate - last recorded maxrate from NM
+ last_avgexemptrate - last recorded avgexemptrate from NM
+ last_maxexemptrate - last recorded maxexemptrate from NM
"""
def __init__(self, xid, name, maxrate, maxexemptrate, bytes, exemptbytes):
self.xid = xid
self.name = name
+ self.time = 0
+ self.exemptbytes = 0
+ self.last_maxrate = default_maxrate
+ self.last_avgrate = default_avgrate
+ self.last_avgexemptrate = default_avgexemptrate
+ self.last_maxexemptrate = default_maxexemptrate
self.reset(maxrate, maxexemptrate, bytes, exemptbytes)
def __repr__(self):
self.bytes = bytes
self.exemptbytes = exemptbytes
+ # If NM except"ns below, and new_max* doesn't get set, use last.
+ new_maxrate = self.last_maxrate
+ new_maxexemptrate = self.last_maxexemptrate
+
# Query Node Manager for max rate overrides
- (new_maxrate, new_maxexemptrate) = nm.query(self.name, [('nm_net_max_rate', -1), ('nm_net_max_exempt_rate', -1)])
- if new_maxrate == -1:
- new_maxrate = default_maxrate
- if new_maxexemptrate == -1:
- new_maxexemptrate = default_maxexemptrate
+ try:
+ vals = nm.query(self.name, [('nm_net_max_rate', self.last_maxrate),
+ ('nm_net_max_exempt_rate', self.last_maxexemptrate),
+ ('nm_net_avg_rate', self.last_avgrate),
+ ('nm_net_avg_exempt_rate', self.last_avgexemptrate)])
+ (new_maxrate, new_maxexemptrate,
+ self.last_avgrate, self.last_avgexemptrate) = vals
+ #If NM is alive, and there is a cap, update new
+ self.last_maxrate = new_maxrate
+ self.last_maxexemptrate = new_maxexemptrate
+
+ except Exception, err:
+ print "Warning: Exception received while querying NM:", err
if new_maxrate != maxrate or new_maxexemptrate != maxexemptrate:
print "%s reset to %s/%s" % \
it will get to send slightly more than the average daily byte
limit.
"""
+
+ # If NM except'ns below, and avg*rate doesn't get set, use last_*.
+ avgrate = self.last_avgrate
+ avgexemptrate = self.last_avgexemptrate
# Query Node Manager for max average rate overrides
- (self.avgrate, self.avgexemptrate) = nm.query(self.name, [('nm_net_avg_rate', -1), ('nm_net_avg_exempt_rate', -1)])
- if self.avgrate == -1:
- self.avgrate = default_avgrate
- if self.avgexemptrate == -1:
- self.avgexemptrate = default_avgexemptrate
-
- # Prepare message parameters from the template
+ try:
+ (avgrate, avgexemptrate) = nm.query(self.name,
+ [('nm_net_avg_rate', self.last_avgrate),
+ ('nm_net_avg_exempt_rate', self.last_avgexemptrate)])
+ #If NM is alive, and there is a cap, update new
+ self.last_avgexemptrate = avgexemptrate
+ self.last_avgrate = avgrate
+ except Exception, err:
+ print "Warning: Exception received while querying NM:", err
+
+ # Prepare message parameters from the template
message = ""
params = {'slice': self.name, 'hostname': socket.gethostname(),
'since': time.asctime(time.gmtime(self.time)) + " GMT",
'date': time.asctime(time.gmtime()) + " GMT",
'period': format_period(period)}
- bytelimit = self.avgrate * period / bits_per_byte
+ bytelimit = avgrate * period / bits_per_byte
if bytes >= (self.bytes + bytelimit) and \
- maxrate > self.avgrate:
- new_maxrate = self.avgrate
+ maxrate > avgrate:
+ new_maxrate = avgrate
else:
new_maxrate = maxrate
params['bytes'] = format_bytes(bytes - self.bytes)
params['maxrate'] = bwlimit.format_tc_rate(maxrate)
params['limit'] = format_bytes(bytelimit)
- params['avgrate'] = bwlimit.format_tc_rate(self.avgrate)
+ params['avgrate'] = bwlimit.format_tc_rate(avgrate)
if verbose:
print "%(slice)s %(class)s " \
- "%(bytes)s/%(limit)s (%(maxrate)s/%(avgrate)s)" % \
+ "%(bytes)s, %(limit)s (%(maxrate)s max/%(avgrate)s avg)" % \
params
# Cap low bandwidth burst rate
message += template % params
print "%(slice)s %(class)s capped at %(avgrate)s (%(bytes)s/%(limit)s)" % params
- exemptbytelimit = self.avgexemptrate * period / bits_per_byte
+ exemptbytelimit = avgexemptrate * period / bits_per_byte
if exemptbytes >= (self.exemptbytes + exemptbytelimit) and \
- maxexemptrate > self.avgexemptrate:
- new_maxexemptrate = self.avgexemptrate
+ maxexemptrate > avgexemptrate:
+ new_maxexemptrate = avgexemptrate
else:
new_maxexemptrate = maxexemptrate
params['bytes'] = format_bytes(exemptbytes - self.exemptbytes)
params['maxrate'] = bwlimit.format_tc_rate(maxexemptrate)
params['limit'] = format_bytes(exemptbytelimit)
- params['avgrate'] = bwlimit.format_tc_rate(self.avgexemptrate)
+ params['avgrate'] = bwlimit.format_tc_rate(avgexemptrate)
if verbose:
print "%(slice)s %(class)s " \
- "%(bytes)s/%(limit)s (%(maxrate)s/%(avgrate)s)" % \
+ "%(bytes)s, %(limit)s (%(maxrate)s max /%(avgrate)s avg)" % \
params
# Cap high bandwidth burst rate
(version, slices) = pickle.load(f)
f.close()
# Check version of data file
- if version != "$Id: bwmon.py,v 1.4 2006/06/02 04:00:00 mlhuang Exp $":
+ if version != "$Id: bwmon.py,v 1.8 2006/07/10 20:22:12 faiyaza Exp $":
print "Not using old version '%s' data file %s" % (version, datafile)
raise Exception
except Exception:
- version = "$Id: bwmon.py,v 1.4 2006/06/02 04:00:00 mlhuang Exp $"
+ version = "$Id: bwmon.py,v 1.8 2006/07/10 20:22:12 faiyaza Exp $"
slices = {}
# Get special slice IDs
root_xid = bwlimit.get_xid("root")
default_xid = bwlimit.get_xid("default")
- # Open connection to Node Manager
+ #Open connection to Node Manager
nm = NM()
live = []
if names and name not in names:
continue
+ #slices is populated from the pickle file
+ #xid is populated from bwlimit (read from /etc/passwd)
if slices.has_key(xid):
slice = slices[xid]
if time.time() >= (slice.time + period) or \
# Mark Huang <mlhuang@cs.princeton.edu>
# Copyright (C) 2006 The Trustees of Princeton University
#
-# $Id: pl_mom.py,v 1.3 2006/05/09 03:22:25 mlhuang Exp $
+# $Id: pl_mom.py,v 1.5 2006/07/10 15:19:35 faiyaza Exp $
#
import os
Simple interface to local Node Manager API
"""
- def __init__(self, url = "http://localhost:812/", timeout = 10):
+ def __init__(self, url = "http://localhost:812/", timeout = 30):
"""
Open a new connection to the local Node Manager
"""
-
socket.setdefaulttimeout(timeout)
try:
self.nm = xmlrpclib.ServerProxy(url)
values = [attribute[1] for attribute in attributes]
if self.nm is not None:
- try:
- # Read rspec (the NM hash code for the slice)
+ # Read rspec (the NM hash code for the slice)
rcap = open("/var/run/pl_nm/%s.vm_rcap" % slice, "r")
rspec = rcap.readline().strip()
rcap.close()
(rc, result) = self.nm.nm_inspect(rspec, attributes)
if rc == 0 and type(result) == list and len(result) == len(values):
values = result
- except Exception, err:
- print "Warning: Exception received while querying Node Manager:", err
-
return values
%define name pl_mom
%define version 0.4
-%define release 3%{?pldistro:.%{pldistro}}%{?date:.%{date}}
+%define release 7%{?pldistro:.%{pldistro}}%{?date:.%{date}}
Summary: PlanetLab node monitoring tools
Name: %{name}
%{_initrddir}/swapmon
/usr/local/planetlab/bin/pl_mop.sh
%{_sysconfdir}/cron.d/pl_mom
+
+%changelog
+* Wed Jul 19 2006 Faiyaz Ahmed <faiyaza@cs.princeton.edu> 0.4-7.planetlab
+- swapmon: Added state info so same reset email does not keep getting
+ sent. If slice is already being killed, will wait for a timeout
+ before trying to kill again.
+
+* Mon Jul 17 2006 Faiyaz Ahmed <faiyaza@cs.princeton.edu> 0.4-6.planetlab
+- swapmon: Added state information per slice. Slices are now killed
+ after consuming more memory than the threshold more than twice.
+ Also fixed the slice restart issue where slices would be restarted
+ and repeated restart emails would be sent.
+
+* Mon Jul 10 2006 Faiyaz Ahmed <faiyaza@cs.princeton.edu> 0.4-5.planetlab
+- bwmon: Initialize default limits
+
+* Mon Jul 10 2006 Faiyaz Ahmed <faiyaza@cs.princeton.edu> 0.4-4.planetlab
+- bwmon: Use cached values if NM stops responding
#
# Mark Huang <mlhuang@cs.princeton.edu>
# Andy Bavier <acb@cs.princeton.edu>
+# Faiyaz Ahmed <faiyaza@cs.princeton.edu>
# Copyright (C) 2004-2006 The Trustees of Princeton University
#
-# $Id: swapmon.py,v 1.4 2006/05/02 17:23:14 mlhuang Exp $
+# $Id: swapmon.py,v 1.9 2006/07/19 19:40:55 faiyaza Exp $
#
import syslog
change_thresh = 5
# Swap utilization at which the largest consumer of physical memory is reset
-reset_thresh = 85
+reset_thresh = 80
# Swap utilization at which the machine is rebooted
reboot_thresh = 95
+# Time to wait before checking slice again after reset
+reset_timeout = 15
+
+# Number of strikes before killing (strike, strike, kill)
+kill_thresh = 2
+
+# Time to wait before removing slice from kill queue (probation)
+kill_timeout = 120
+
+# Don't email the same message more than once in the same emailtimeout interval
+email_timeout = 1800
+
# Minimum physical memory utilization to be considered the largest consumer
min_thresh = 10
%(date)s %(hostname)s alarm %(slice)s
""".lstrip()
+# Message sent after a slice has been killed
+kill_subject = "pl_mom killed slice %(slice)s on %(hostname)s"
+kill_body = \
+"""
+Sometime before %(date)s, swap space was
+nearly exhausted on %(hostname)s.
+
+Slice %(slice)s was killed since it was the largest consumer of
+physical memory at %(rss)s (%(percent)4.1f%%) after repeated restarts.
+
+Please reply to this message explaining the nature of your experiment,
+and what you are doing to address the problem.
+
+%(slice)s processes prior to reset:
+
+%(table)s
+
+%(date)s %(hostname)s reset %(slice)s
+""".lstrip()
+
+
+
+class Reset:
+ """
+ Keeps track of state information for resets and kills
+
+ resettimeleft - timeout before checking for next reset
+ resetcount - number of strikes
+ killtimeleft - time out before removing from kill queue
+ {kill,reset}mail - Time of last email
+ kill - State of kill. If slice is already being killed, wait before retry.
+ """
+
+ def __init__(self,name):
+ self.name = name
+ self.resettimeleft = reset_timeout
+ self.resetcount = 0
+ self.resetmail = 0
+ self.kill = False
+ self.killtimeleft = kill_timeout
+ self.killmail = 0
+
+ def __repr__(self):
+ return self.name
+
+ def update(self):
+ # Count down for next check of reset slice.
+ if self.resettimeleft > 0:
+ self.resettimeleft -= 1
+ if debug and verbose:
+ print "%s has %s seconds in probation" \
+ %(self.name, self.killtimeleft)
+ if self.killtimeleft > 0:
+ # Count down kill probation timer (killtimeleft)
+ self.killtimeleft -= 1
+ if self.killtimeleft == 1:
+ print "%s is out of probation" % self.name
+ else:
+ # Once out of probation period (killtimeleft), remove strikes
+ self.resetcount = 0
+ self.kill = False
+
+
+ # Check to see if a slice needs to be killed. If it has rules more than kill_thresh in
+ # the probation period (kill_timeout) send an email, kill the slice.
+ def checkkill(self,params):
+ if self.killtimeleft > 0 and self.resetcount >= kill_thresh and \
+ self.kill == False:
+ self.kill = True
+ if debug:
+ print kill_subject % params
+ print kill_body % params
+ try:
+ pid = os.fork()
+ if pid == 0:
+ print "Slice %s is being killed." % self.name
+ vserver = VServer(self.name)
+ vserver.stop()
+ os._exit(0)
+ else:
+ os.waitpid(pid,0)
+ except Exception, err:
+ print "Warning: Exception received while killing slice %s: %s" \
+ % self.name, err
+ if (time.time() - self.killmail) > email_timeout:
+ slicemail(self.name, kill_subject % params, kill_body % params)
+ print "Sending KILL email for slice %s" % self.name
+ self.killmail = time.time()
+ return True
+ return False
+
+ # Reset slice after checking to see if slice is out of timeout.
+ # Increment resetcount, check to see if larger than kill_thresh.
+ def reset(self, params):
+ # If its the first reset or if its been reset before
+ # and we are out of the reset timeout.
+ if self.resetcount == 0 or self.resettimeleft == 0:
+ # Do we need to kill this slice? Check history first.
+ if self.checkkill(params):
+ return
+ # Update counters
+ self.resetcount += 1
+ self.killtimeleft = kill_timeout
+ self.resettimeleft = reset_timeout
+ print "%s has %s seconds to die and has been reset %s times" \
+ %(self.name, self.resettimeleft, self.resetcount)
+ if debug:
+ print reset_subject % params
+ print reset_body % params
+ try:
+ pid = os.fork()
+ if pid == 0:
+ print "Resetting slice " + self.name
+ vserver = VServer(self.name)
+ vserver.stop()
+ vserver.start(wait = False)
+ os._exit(0)
+ else:
+ os.waitpid(pid,0)
+ except Exception, err:
+ print "Warning: Exception received while resetting slice %s:" \
+ % self.name, err
+ if (time.time() - self.resetmail) > email_timeout:
+ slicemail(self.name, reset_subject % params, reset_body % params)
+ print "Sending Reset email for slice %s" % self.name
+ self.resetmail = time.time()
+
+
def usage():
print """
Usage: %s [OPTIONS]...
(version, slices) = pickle.load(f)
f.close()
# Check version of data file
- if version != "$Id: swapmon.py,v 1.4 2006/05/02 17:23:14 mlhuang Exp $":
+ if version != "$Id: swapmon.py,v 1.9 2006/07/19 19:40:55 faiyaza Exp $":
print "Not using old version '%s' data file %s" % (version, datafile)
raise Exception
# Delete data file
os.unlink(datafile)
except Exception:
- version = "$Id: swapmon.py,v 1.4 2006/05/02 17:23:14 mlhuang Exp $"
+ version = "$Id: swapmon.py,v 1.9 2006/07/19 19:40:55 faiyaza Exp $"
slices = {}
# Query process table every 30 seconds, or when a large change in
# System slices that we have warned but could not reset
warned = []
+ # Slices that were reset
+ resetlist = {}
+
while True:
used = swap_used()
+
+ for resetslice in resetlist.keys():
+ resetlist[resetslice].update()
+
if last_used is None:
last_used = used
- if verbose:
+ if verbose:
print "%d%% swap consumed" % used
if used >= reboot_thresh:
slicelist.sort(lambda a, b: b['rss'] - a['rss'])
for slice in slicelist:
percent = 100. * slice['rss'] / total_rss
+
if percent < min_thresh:
continue
-
+
print "%d%% swap consumed, slice %s is using %s (%d%%) of memory" % \
(used,
slice['name'],
# Match slice name against system slice patterns
is_system_slice = filter(None, [re.match(pattern, slice['name']) for pattern in system_slices])
- if is_system_slice:
- # Do not reset system slices, just warn once
- if slice['name'] not in warned:
- warned.append(slice['name'])
- if debug:
- print alarm_subject % params
- print alarm_body % params
- else:
- print "Warning slice " + slice['name']
- slicemail(slice['name'], alarm_subject % params, alarm_body % params)
+ if is_system_slice:
+ if slice['name'] not in warned:
+ warned.append(slice['name'])
+ if debug:
+ print alarm_subject % params
+ print alarm_body % params
+ else:
+ print "Warning slice " + slice['name']
+ slicemail(slice['name'], alarm_subject % params,
+ alarm_body % params)
else:
- # Otherwise, reset
- if debug:
- print reset_subject % params
- print reset_body % params
- else:
- try:
- pid = os.fork()
- if pid == 0:
- print "Resetting slice " + slice['name']
- vserver = VServer(slice['name'])
- vserver.stop()
- vserver.start(wait = False)
- os._exit(0)
- else:
- os.waitpid(pid, 0)
- except Exception, err:
- print "Warning: Exception received while resetting slice %s:" % slice['name'], err
- slicemail(slice['name'], reset_subject % params, reset_body % params)
- break
+ # Reset slice
+ if not resetlist.has_key(slice['name']):
+ resetlist[slice['name']] = Reset(slice['name'])
+ resetlist[slice['name']].reset(params)
elif timer <= 0 or used >= (last_used + change_thresh):
if used >= (last_used + change_thresh):
git://git.onelab.eu / mom.git / commitdiff