# Faiyaz Ahmed <faiyaza@cs.princeton.edu>
# Copyright (C) 2004-2006 The Trustees of Princeton University
#
-# $Id: swapmon.py,v 1.5 2006/05/09 03:23:57 mlhuang Exp $
+# $Id: swapmon.py,v 1.10 2006/08/16 16:18:45 faiyaza Exp $
#
import syslog
change_thresh = 5
# Swap utilization at which the largest consumer of physical memory is reset
-reset_thresh = 75
+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 = 10
+reset_timeout = 15
# Number of strikes before killing (strike, strike, kill)
kill_thresh = 2
-# Time to wait before removing slice from kill queue
-kill_timeout = 30
+# Time to wait before removing slice from kill queue (probation)
+kill_timeout = 120
-# Minimum physical memory utilization to be considered the largest consumer
-min_thresh = 10
+# Don't email the same message more than once in the same emailtimeout interval
+email_timeout = 1800
+
+# Physical size threshold to be considered a consumer. Rationale is if there are no procs
+# with a size at least as large as this, then there is a slow leaker; better to just reboot.
+rss_min = 150 * 1024
# System slices that should not be reset (regexps)
system_slices = ['root', PLC_SLICE_PREFIX + '_']
nearly exhausted on %(hostname)s.
Slice %(slice)s was reset since it was the largest consumer of
-physical memory at %(rss)s (%(percent)4.1f%%).
+physical memory at %(rss)s (%(percent)4.1f%%) (%(sz)s writable).
Please reply to this message explaining the nature of your experiment,
and what you are doing to address the problem.
nearly exhausted on %(hostname)s.
System slice %(slice)s was the largest consumer of physical memory at
-%(rss)s (%(percent)4.1f%%). It was not reset, but please verify its
-behavior.
+%(rss)s (%(percent)4.1f%%) (%(sz)s writable). It was not reset,
+but please verify its behavior.
%(slice)s processes prior to alarm:
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.
+physical memory at %(rss)s (%(percent)4.1f%%) (%(sz)s writable)
+after repeated restarts.
Please reply to this message explaining the nature of your experiment,
and what you are doing to address the problem.
resettimeleft - timeout before checking for next reset
resetcount - number of strikes
- killtimeleft - time out before removing from kill queue
+ 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.killtimeleft = kill_timeout
+ self.killmail = 0
def __repr__(self):
return self.name
if self.resettimeleft > 0:
self.resettimeleft -= 1
if debug and verbose:
- print "%s has %s seconds to die and has been reset %s times" \
- %(self.name, self.resettimeleft, self.resetcount)
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 strike
+ # Once out of probation period (killtimeleft), remove strikes
self.resetcount = 0
- # 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.
+ # Check to see if a slice needs to be killed. If it has been killed 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:
if debug:
pid = os.fork()
if pid == 0:
print "Slice %s is being killed." % self.name
- vserver = VServer(self.name)
- vserver.stop()
+ 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
- slicemail(self.name, kill_subject % params, kill_body % params)
+ 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
+ # If its the first reset (came back after kill)
+ # 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.
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)
+ 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:" \
+ print "Warning: Exception received while resetting slice %s:" \
% self.name, err
- slicemail(self.name, reset_subject % params, reset_body % params)
-
+ 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():
def slicestat(names = None):
"""
Get status of specified slices (if names is None or empty, all
- slices). vsize and rss are in KiB. Returns
+ slices). vsize, sz, and rss are in KiB. Returns
{xid: {'xid': slice_id,
'name': slice_name,
'procs': [{'pid': pid, 'xid': slice_id, 'user', username, 'cmd': command,
- 'vsize': virtual_kib, 'rss': physical_kib,
+ 'vsize': virtual_kib, 'sz': potential_kib, 'rss': physical_kib,
'pcpu': cpu_percent, 'pmem': mem_percent}]
'vsize': total_virtual_kib,
+ 'sz': total_potential_kib,
'rss': total_physical_kib}}
"""
# Mandatory fields. xid is a virtual field inserted by vps. Make
# sure cmd is last so that it does not get truncated
# automatically.
- fields = ['pid', 'xid', 'user', 'vsize', 'rss', 'pcpu', 'pmem', 'cmd']
+ fields = ['pid', 'xid', 'user', 'vsize', 'sz', 'rss', 'pcpu', 'pmem', 'cmd']
# vps inserts xid after pid in the output, but ps doesn't know
# what the field means.
if slices.has_key(proc['xid']):
slice = slices[proc['xid']]
else:
- slice = {'xid': proc['xid'], 'name': name, 'procs': [], 'vsize': 0, 'rss': 0}
+ slice = {'xid': proc['xid'], 'name': name, 'procs': [], 'vsize': 0, 'sz': 0, 'rss': 0}
slice['procs'].append(proc)
slice['vsize'] += proc['vsize']
+ slice['sz'] += proc['sz']
slice['rss'] += proc['rss']
slices[proc['xid']] = slice
def memtotal():
"""
- Returns total physical memory on the system in KiB.
+ Returns total physical and swap memory on the system in KiB.
"""
+ mem = 0
+ swap = 0
+
meminfo = open("/proc/meminfo", "r")
- line = meminfo.readline()
+ for line in meminfo.readlines():
+ try:
+ (name, value, kb) = line.split()
+ except:
+ continue
+ if name == "MemTotal:":
+ mem = int(value)
+ elif name == "SwapTotal:":
+ swap = int(value)
meminfo.close()
- if line[0:8] == "MemTotal":
- # MemTotal: 255396 kB
- (name, value, kb) = line.split()
- return int(value)
- return 0
+ return (mem, swap)
def swap_used():
"""
return 100 * total_used / total_swap
-def summary(names = None, total_rss = memtotal()):
+def summary(slices = None, total_mem = None, total_swap = None):
"""
Return a summary of memory usage by slice.
"""
- slicelist = slicestat(names).values()
- slicelist.sort(lambda a, b: b['rss'] - a['rss'])
-
- table = "%-20s%10s%24s\n\n" % ("Slice", "Processes", "Memory Usage")
+ if not slices:
+ slices = slicestat()
+ slicelist = slices.values()
+ slicelist.sort(lambda a, b: b['sz'] - a['sz'])
+ if total_mem is None or total_swap is None:
+ (total_mem, total_swap) = memtotal()
+
+ table = "%-20s%10s%24s%24s\n\n" % ("Slice", "Processes", "Memory Usage", "Potential Usage")
for slice in slicelist:
- table += "%-20s%10d%16s (%4.1f%%)\n" % \
+ table += "%-20s%10d%16s (%4.1f%%)%16s (%4.1f%%)\n" % \
(slice['name'], len(slice['procs']),
format_bytes(slice['rss'] * 1024, si = False),
- 100. * slice['rss'] / total_rss)
+ 100. * slice['rss'] / total_mem,
+ format_bytes(slice['sz'] * 1024, si = False),
+ 100. * slice['sz'] / (total_mem + total_swap))
+
return table
def main():
# Defaults
global debug, verbose, datafile
- global period, change_thresh, reset_thresh, reboot_thresh, min_thresh, system_slices
+ global period, change_thresh, reset_thresh, reboot_thresh, rss_min, system_slices
# All slices
names = []
elif opt == "--reboot-thresh":
reboot_thresh = int(optval)
elif opt == "--min-thresh":
- min_thresh = int(optval)
+ rss_min = int(optval)
elif opt == "--system-slice":
system_slices.append(optval)
elif opt == "--status":
- print summary(names)
+ print summary(slicestat(names))
sys.exit(0)
else:
usage()
syslog.openlog("swapmon")
sys.stdout = sys.stderr = Logger()
- # Get total physical memory
- total_rss = memtotal()
+ # Get total memory
+ (total_mem, total_swap) = memtotal()
try:
f = open(datafile, "r+")
(version, slices) = pickle.load(f)
f.close()
# Check version of data file
- if version != "$Id: swapmon.py,v 1.5 2006/05/09 03:23:57 mlhuang Exp $":
+ if version != "$Id: swapmon.py,v 1.10 2006/08/16 16:18:45 faiyaza Exp $":
print "Not using old version '%s' data file %s" % (version, datafile)
raise Exception
params = {'hostname': socket.gethostname(),
'date': time.asctime(time.gmtime()) + " GMT",
- 'table': summary(total_rss)}
+ 'table': summary(slices, total_mem, total_swap)}
if debug:
print rebooted_subject % params
# Delete data file
os.unlink(datafile)
except Exception:
- version = "$Id: swapmon.py,v 1.5 2006/05/09 03:23:57 mlhuang Exp $"
+ version = "$Id: swapmon.py,v 1.10 2006/08/16 16:18:45 faiyaza Exp $"
slices = {}
# Query process table every 30 seconds, or when a large change in
used = swap_used()
for resetslice in resetlist.keys():
- resetlist[resetslice].update()
+ resetlist[resetslice].update()
if last_used is None:
last_used = used
+
if verbose:
print "%d%% swap consumed" % used
bwlimit.run("/bin/sync; /sbin/reboot -f")
elif used >= reset_thresh:
+ if debug:
+ print "Memory used = %s" %(used)
# Try and find a hog
slicelist = slices.values()
slicelist.sort(lambda a, b: b['rss'] - a['rss'])
for slice in slicelist:
- percent = 100. * slice['rss'] / total_rss
+ percent = 100. * slice['rss'] / total_mem
- if percent < min_thresh:
+ if slice['rss'] < rss_min:
continue
print "%d%% swap consumed, slice %s is using %s (%d%%) of memory" % \
slice['procs'].sort(lambda a, b: b['rss'] - a['rss'])
- table = "%5s %10s %10s %4s %4s %s\n\n" % ("PID", "VIRT", "RES", '%CPU', '%MEM', 'COMMAND')
+ table = "%5s %10s %10s %10s %4s %4s %s\n\n" % ("PID", "VIRT", "SZ", "RES", '%CPU', '%MEM', 'COMMAND')
for proc in slice['procs']:
- table += "%5s %10s %10s %4.1f %4.1f %s\n" % \
+ table += "%5s %10s %10s %10s %4.1f %4.1f %s\n" % \
(proc['pid'],
format_bytes(proc['vsize'] * 1024, si = False),
+ format_bytes(proc['sz'] * 1024, si = False),
format_bytes(proc['rss'] * 1024, si = False),
proc['pcpu'], proc['pmem'], proc['cmd'])
'table': table,
'slice': slice['name'],
'rss': format_bytes(slice['rss'] * 1024, si = False),
+ 'sz': format_bytes(slice['sz'] * 1024, si = False),
'percent': percent}
# 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:
+ # 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)
else:
- # Reset slice
- if not resetlist.has_key(slice['name']):
- resetlist[slice['name']] = Reset(slice['name'])
- resetlist[slice['name']].reset(params)
+ # Reset slice
+ if not resetlist.has_key(slice['name']):
+ resetlist[slice['name']] = Reset(slice['name'])
+ resetlist[slice['name']].reset(params)
+ slices = slicestat(names)
- elif timer <= 0 or used >= (last_used + change_thresh):
+ if timer <= 0 or used >= (last_used + change_thresh):
if used >= (last_used + change_thresh):
print "%d%% swap consumed, %d%% in last %d seconds" % \
(used, used - last_used, period - timer)
git://git.onelab.eu / mom.git / blobdiff