# Faiyaz Ahmed <faiyaza@cs.princeton.edu>
# Copyright (C) 2004-2006 The Trustees of Princeton University
#
-# $Id: swapmon.py,v 1.11 2006/12/02 19:11:47 mlhuang Exp $
+# $Id$
#
import syslog
# Defaults
debug = False
verbose = 0
-datafile = "/var/lib/misc/swapmon.dat"
-
+DATAFILE = "/var/lib/misc/swapmon.dat"
+VERSION = "$Id$"
# Seconds between process analysis
period = 30
reboot_thresh = 95
# Time to wait before checking slice again after reset
-reset_timeout = 15
+reset_timeout = 25
# Number of strikes before killing (strike, strike, kill)
kill_thresh = 2
Please reply to this message explaining the nature of your experiment,
and what you are doing to address the problem.
+http://summer.cs.princeton.edu/status/tabulator.cgi?table=slices/table_%(slice)s
+
%(slice)s processes prior to reset:
%(table)s
def checkkill(self,params):
if self.killtimeleft > 0 and self.resetcount >= kill_thresh:
if debug:
- print kill_subject % params
- print kill_body % params
+ print kill_subject % params
+ print kill_body % params
try:
pid = os.fork()
if pid == 0:
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:" \
+ 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)
--system-slice=SLICE System slice that should not be reset
--status Print memory usage statistics and exit
-h, --help This message
-""".lstrip() % (sys.argv[0], debug, verbose, datafile, format_period(period))
+""".lstrip() % (sys.argv[0], debug, verbose, DATAFILE, format_period(period))
def slicestat(names = None):
"""
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,
+ 'name': slice_name,
+ 'procs': [{'pid': pid, 'xid': slice_id, 'user', username, 'cmd': command,
'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}}
+ 'vsize': total_virtual_kib,
+ 'sz': total_potential_kib,
+ 'rss': total_physical_kib}}
"""
# Mandatory fields. xid is a virtual field inserted by vps. Make
"""
Returns total physical and swap memory on the system in KiB.
"""
-
mem = 0
swap = 0
-
meminfo = open("/proc/meminfo", "r")
for line in meminfo.readlines():
try:
elif name == "SwapTotal:":
swap = int(value)
meminfo.close()
-
return (mem, swap)
def swap_used():
"""
Returns swap utilization on the system as a whole percentage (0-100).
"""
-
total_swap = 0
total_used = 0
-
try:
swaps = open("/proc/swaps", "r")
# Eat header line
total_used += int(used)
except ValueEror, err:
pass
- except (IOError, KeyError), err:
- pass
+ except (IOError, KeyError), err: pass
- return 100 * total_used / total_swap
+ swapused = 100 * total_used / total_swap
+ if debug: print "%s percent swap used" % swapused
+ return swapused
def summary(slices = None, total_mem = None, total_swap = None):
"""
Return a summary of memory usage by slice.
"""
- if not slices:
- slices = slicestat()
+ 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:
100. * slice['rss'] / total_mem,
format_bytes(slice['sz'] * 1024, si = False),
100. * slice['sz'] / (total_mem + total_swap))
-
-
return table
+def formtable(slice, percent):
+ '''
+ Makes pretty message to email with human readable ps values.
+ '''
+ 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 %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'])
+
+ prettytable = {'hostname': socket.gethostname(),
+ 'date': time.asctime(time.gmtime()) + " GMT",
+ 'table': table,
+ 'slice': slice['name'],
+ 'rss': format_bytes(slice['rss'] * 1024, si = False),
+ 'sz': format_bytes(slice['sz'] * 1024, si = False),
+ 'percent': percent}
+ return prettytable
+
+def readdat():
+ '''
+ Return dictionary of vps (slicestat) from datfile left behind by OOM
+ before rebooting. If none file, just grab the latest dict (slicestat)
+ and return that. If dat file found, means we rebooted, send an email to
+ pl_mom@pl.
+ '''
+ try:
+ f = open(DATAFILE, "r+")
+ if verbose:
+ print "Loading %s" % DATAFILE
+ (v, slices) = pickle.load(f)
+ f.close()
+ # Check version of data file
+ if v != VERSION:
+ print "Not using old version '%s' data file %s" % (v, DATAFILE)
+ raise Exception
+
+ params = {'hostname': socket.gethostname(),
+ 'date': time.asctime(time.gmtime()) + " GMT",
+ 'table': summary(slices, total_mem, total_swap)}
+ if debug:
+ print rebooted_subject % params
+ print rebooted_body % params
+ else:
+ slicemail(None, rebooted_subject % params, rebooted_body % params)
+
+ # Delete data file
+ os.unlink(DATAFILE)
+ except Exception:
+ slices = slicestat()
+
+ return slices
+
+
+def writedat(slices):
+ """
+ Write (slices) to pickled datfile.
+ """
+ if verbose: print "Saving %s" % DATAFILE
+ f = open(DATAFILE, "w")
+ pickle.dump((VERSION, slices), f)
+ f.close()
+
+
def main():
# Defaults
- global debug, verbose, datafile
+ global debug, verbose, DATAFILE, VERSION
global period, change_thresh, reset_thresh, reboot_thresh, rss_min, system_slices
# All slices
names = []
elif opt == "-v" or opt == "--verbose":
verbose += 1
elif opt == "-f" or opt == "--file":
- datafile = optval
+ DATAFILE = optval
elif opt == "-s" or opt == "--slice":
names.append(optval)
elif opt == "-p" or opt == "--period":
# Get total memory
(total_mem, total_swap) = memtotal()
-
- try:
- f = open(datafile, "r+")
- if verbose:
- print "Loading %s" % datafile
- (version, slices) = pickle.load(f)
- f.close()
- # Check version of data file
- if version != "$Id: swapmon.py,v 1.11 2006/12/02 19:11:47 mlhuang Exp $":
- print "Not using old version '%s' data file %s" % (version, datafile)
- raise Exception
-
- params = {'hostname': socket.gethostname(),
- 'date': time.asctime(time.gmtime()) + " GMT",
- 'table': summary(slices, total_mem, total_swap)}
-
- if debug:
- print rebooted_subject % params
- print rebooted_body % params
- else:
- slicemail(None, rebooted_subject % params, rebooted_body % params)
-
- # Delete data file
- os.unlink(datafile)
- except Exception:
- version = "$Id: swapmon.py,v 1.11 2006/12/02 19:11:47 mlhuang Exp $"
- slices = {}
+ slices = readdat()
# Query process table every 30 seconds, or when a large change in
# swap utilization is detected.
while True:
used = swap_used()
-
- for resetslice in resetlist.keys():
- resetlist[resetslice].update()
+ if last_used is None: last_used = used
+
+ # If we've reset you recently, update timers.
+ for resetslice in resetlist.keys():
+ resetlist[resetslice].update()
+ # If you've been good, remove you from our list.
+ if resetlist[resetslice].killtimeleft == 0 and \
+ resetlist[resetslice].resettimeleft == 0:
+ del resetlist[resetslice]
+
+ if verbose: print "%d%% swap consumed" % used
+
+ if used >= reboot_thresh:
+ # Dump slice state before rebooting
+ writedat(slices)
+ # Goodbye, cruel world
+ print "%d%% swap consumed, rebooting" % used
+ if not debug: bwlimit.run("/bin/sync; /sbin/reboot -f")
+ elif used >= reset_thresh:
+ # Try and find a hog
+ slicelist = slices.values()
+ # Puts largest on top.
+ slicelist.sort(lambda a, b: b['rss'] - a['rss'])
+ for slice in slicelist:
+ percent = 100. * slice['rss'] / total_mem
+ if slice['rss'] < rss_min: continue
+ print "%d%% swap consumed, slice %s is using %s (%d%%) of memory" % \
+ (used,
+ slice['name'],
+ format_bytes(slice['rss'] * 1024, si = False),
+ percent)
+ slice['procs'].sort(lambda a, b: b['rss'] - a['rss'])
+ # Make a pretty table.
+ params = formtable(slice, percent)
+ # Match slice name against system slice patterns
+ is_system_slice = filter(None,
+ [re.match(pattern, slice['name']) for pattern in system_slices])
- if last_used is None:
- last_used = used
-
- if verbose:
- print "%d%% swap consumed" % used
-
- if used >= reboot_thresh:
- # Dump slice state before rebooting
- if verbose: print "Saving %s" % datafile
- f = open(datafile, "w")
- pickle.dump((version, slices), f)
- f.close()
-
- # Goodbye, cruel world
- print "%d%% swap consumed, rebooting" % used
- if not debug: 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_mem
- if slice['rss'] < rss_min: continue
- print "%d%% swap consumed, slice %s is using %s (%d%%) of memory" % \
- (used,
- slice['name'],
- format_bytes(slice['rss'] * 1024, si = False),
- percent)
-
- slice['procs'].sort(lambda a, b: b['rss'] - a['rss'])
- 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 %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'])
-
- params = {'hostname': socket.gethostname(),
- 'date': time.asctime(time.gmtime()) + " GMT",
- '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
+ if is_system_slice:
+ if slice['name'] not in warned:
+ warned.append(slice['name'])
+ print "Warning slice " + slice['name']
+ if debug:
+ print alarm_subject % params
+ print alarm_body % params
+ else:
+ slicemail(slice['name'], alarm_subject % params,
+ 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)
- slices = slicestat(names)
+ # Reset slice
+ if not resetlist.has_key(slice['name']):
+ resetlist[slice['name']] = Reset(slice['name'])
+ resetlist[slice['name']].reset(params)
+ # wait period vefore recalculating swap. If in danger, recalc.
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)
+ (used, used - last_used, period - timer)
# Get slice state
slices = slicestat(names)
# Reset timer
timer = period
# Keep track of large changes in swap utilization
last_used = used
-
timer -= 1
time.sleep(1)