# Faiyaz Ahmed <faiyaza@cs.princeton.edu>
# Copyright (C) 2004-2006 The Trustees of Princeton University
#
-# $Id: bwmon.py,v 1.3.2.5 2006/08/24 20:16:14 mlhuang Exp $
+# $Id: bwmon.py,v 1.15 2006/12/13 21:39:23 faiyaza Exp $
#
import syslog
datafile = "/var/lib/misc/bwmon.dat"
nm = None
+# Burst to line rate (or node cap). Set by NM.
default_maxrate = bwlimit.get_bwcap()
-
default_maxexemptrate = bwlimit.bwmax
+# What we cap to when slices break the rules.
# 500 Kbit or 5.4 GB per day
-default_avgrate = 500000
-
+#default_avgrate = 500000
# 1.5 Mbit or 16.4 GB per day
-default_avgexemptrate = 1500000
+#default_avgexemptrate = 1500000
+
+# 5.4 Gbyte per day. 5.4 * 1024 k * 1024M * 1024G
+# 5.4 Gbyte per day max allowed transfered per recording period
+default_ByteMax = 5798205850
+default_ByteThresh = int(.8 * default_ByteMax)
+# 16.4 Gbyte per day max allowed transfered per recording period to I2
+default_ExemptByteMax = 17609365914
+default_ExemptByteThresh = int(.8 * default_ExemptByteMax)
+
# Average over 1 day
period = 1 * seconds_per_day
%(hostname)s to %(class)s destinations
since %(since)s.
-Its maximum %(class)s burst rate will be capped at %(avgrate)s
+Its maximum %(class)s burst rate will be capped at %(new_maxrate)s
until %(until)s.
Please reduce the average %(class)s transmission rate
-of the slice to %(avgrate)s, or %(limit)s per %(period)s.
+of the slice %(limit)s per %(period)s.
""".lstrip()
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 (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
+ ByteMax - total volume of data allowed
+ ByteThresh - After thresh, cap node to (maxbyte - bytes)/(time left in period)
+ ExemptByteMax - Same as above, but for i2.
+ ExemptByteThresh - i2 ByteThresh
+ maxrate - max_rate slice attribute.
+ maxexemptrate - max_exempt_rate slice attribute.
+ self.emailed = did we email during this recording period
+
"""
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.time = 0
+ self.bytes = 0
+ self.exemptbytes = 0
+ self.ByteMax = default_ByteMax
+ self.ByteThresh = default_ByteThresh
+ self.ExemptByteMax = default_ExemptByteMax
+ self.ExemptByteThresh = default_ExemptByteThresh
+ self.maxrate = default_maxrate
+ self.maxexemptrate = default_maxexemptrate
+ self.emailed = False
+
+ # Get real values where applicable
self.reset(maxrate, maxexemptrate, bytes, exemptbytes)
def __repr__(self):
return self.name
+ def updateSliceAttributes(self):
+ # Query Node Manager for max rate overrides
+ try:
+ vals = nm.query(self.name,
+ [('nm_net_max_rate', self.maxrate),
+ ('nm_net_max_exempt_rate', self.maxexemptrate),
+ ("nm_net_max_byte", int(self.ByteMax / 1024)),
+ ("nm_net_max_exempt_byte", int(self.ExemptByteMax / 1024)),
+ ("nm_net_max_thresh_byte", int( .8 * self.ByteMax / 1024)),
+ ("nm_net_max_thresh_exempt_byte", int(.8 * self.ExemptByteMax / 1024)),
+ ("nm_net_avg_rate", 0),
+ ("nm_net_avg_exempt_rate", 0)])
+
+ (self.maxrate,
+ self.maxexemptrate,
+ ByteMax,
+ ExemptByteMax,
+ ByteThresh,
+ ExemptByteThresh,
+ avgrate,
+ avgexemptrate) = vals
+
+ # The shitty bit. Gotta bias the limits so as not to overflow xmlrpc
+ self.ByteMax = ByteMax * 1024
+ self.ByteThresh = ByteThresh * 1024
+ self.ExemptByteMax = ExemptByteMax * 1024
+ self.ExemptByteThresh = ExemptByteThresh * 1024
+
+ # The hack here is that when i pass 0 to the xmlrpc request to NM,
+ # for rate limits and it comes back non zero, then screw the byte limits.
+ # Mult by the period and recompute the byte limits. The thought is
+ # If/when PLC switches to byte limits, the avgrates wont be used as
+ # slice attributes and will return as 0
+ if (avgrate != 0):
+ self.ByteMax = avgrate * period
+ self.ByteThresh = int(self.ByteMax * .8)
+
+ if (avgexemptrate != 0):
+ self.ExemptByteMax = avgexemptrate * period
+ self.ExemptByteThresh = int(self.ExemptByteMax * .8)
+
+ except Exception, err:
+ print "Warning: Exception received while querying NM:", err
+
def reset(self, maxrate, maxexemptrate, bytes, exemptbytes):
"""
Begin a new recording period. Remove caps by restoring limits
to their default values.
"""
+
+ # Query Node Manager for max rate overrides
+ self.updateSliceAttributes()
# Reset baseline time
self.time = time.time()
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
- 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
+ # Reset email
+ self.emailed = False
- if new_maxrate != maxrate or new_maxexemptrate != maxexemptrate:
+ if (self.maxrate != maxrate) or (self.maxexemptrate != maxexemptrate):
print "%s reset to %s/%s" % \
(self.name,
- bwlimit.format_tc_rate(new_maxrate),
- bwlimit.format_tc_rate(new_maxexemptrate))
- bwlimit.set(xid = self.xid, maxrate = new_maxrate, maxexemptrate = new_maxexemptrate)
+ bwlimit.format_tc_rate(self.maxrate),
+ bwlimit.format_tc_rate(self.maxexemptrate))
+ bwlimit.set(xid = self.xid, maxrate = self.maxrate, maxexemptrate = self.maxexemptrate)
def update(self, maxrate, maxexemptrate, bytes, exemptbytes):
"""
- Update byte counts and check if average rates have been
- exceeded. In the worst case (instantaneous usage of the entire
- average daily byte limit at the beginning of the recording
- period), the slice will be immediately capped and will get to
- send twice the average daily byte limit. In the common case,
- it will get to send slightly more than the average daily byte
- limit.
+ Update byte counts and check if byte limits have been
+ exceeded.
"""
-
- # 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
- 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
+
+ # Query Node Manager for max rate overrides
+ self.updateSliceAttributes()
+
+ if verbose:
+ print("\n%s slice attributes "\
+ "maxrate %s, maxexemptrate %s" % \
+ (self.name,
+ bwlimit.format_tc_rate(maxrate),
+ bwlimit.format_tc_rate(maxexemptrate)))
+
+ # 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 = avgrate * period / bits_per_byte
- if bytes >= (self.bytes + bytelimit) and \
- maxrate > avgrate:
- new_maxrate = avgrate
+ if bytes >= (self.bytes + self.ByteThresh):
+ new_maxrate = \
+ int((self.ByteMax - self.bytes + bytes)/(period - time.time() - self.time))
else:
new_maxrate = maxrate
params['class'] = "low bandwidth"
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(avgrate)
+ params['limit'] = format_bytes(self.ByteMax)
+ params['new_maxrate'] = bwlimit.format_tc_rate(new_maxrate)
if verbose:
print "%(slice)s %(class)s " \
- "%(bytes)s, %(limit)s (%(maxrate)s max/%(avgrate)s avg)" % \
+ "%(bytes)s of %(limit)s (%(new_maxrate)s maxrate)" % \
params
# Cap low bandwidth burst rate
if new_maxrate != maxrate:
message += template % params
- print "%(slice)s %(class)s capped at %(avgrate)s (%(bytes)s/%(limit)s)" % params
-
- exemptbytelimit = avgexemptrate * period / bits_per_byte
- if exemptbytes >= (self.exemptbytes + exemptbytelimit) and \
- maxexemptrate > avgexemptrate:
- new_maxexemptrate = avgexemptrate
+ print "%(slice)s %(class)s capped at %(new_maxrate)s " % params
+
+ if exemptbytes >= (self.exemptbytes + self.ExemptByteThresh):
+ new_maxexemptrate = \
+ int((self.ExemptByteMax - (self.bytes + bytes))/(period - (time.time() - self.time)))
else:
new_maxexemptrate = maxexemptrate
params['class'] = "high bandwidth"
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(avgexemptrate)
+ params['limit'] = format_bytes(self.ExemptByteMax)
+ params['new_maxexemptrate'] = bwlimit.format_tc_rate(new_maxexemptrate)
if verbose:
print "%(slice)s %(class)s " \
- "%(bytes)s, %(limit)s (%(maxrate)s max /%(avgrate)s avg)" % \
- params
+ "%(bytes)s of %(limit)s (%(new_maxrate)s maxrate)" % params
# Cap high bandwidth burst rate
if new_maxexemptrate != maxexemptrate:
message += template % params
- print "%(slice)s %(class)s capped at %(avgrate)s (%(bytes)s/%(limit)s)" % params
+ print "%(slice)s %(class)s capped at %(new_maxexemptrate)s" % params
# Apply parameters
if new_maxrate != maxrate or new_maxexemptrate != maxexemptrate:
bwlimit.set(xid = self.xid, maxrate = new_maxrate, maxexemptrate = new_maxexemptrate)
# Notify slice
- if message:
+ if message and self.emailed == False:
subject = "pl_mom capped bandwidth of slice %(slice)s on %(hostname)s" % params
if debug:
print subject
print message + (footer % params)
else:
+ self.emailed = True
slicemail(self.name, subject, message + (footer % params))
+
+
def usage():
print """
Usage: %s [OPTIONS]...
Options:
- -d, --debug Enable debugging (default: %s)
- -v, --verbose Increase verbosity level (default: %d)
- -f, --file=FILE Data file (default: %s)
- -s, --slice=SLICE Constrain monitoring to these slices (default: all)
+ -d, --debug Enable debugging (default: %s)
+ -v, --verbose Increase verbosity level (default: %d)
+ -f, --file=FILE Data file (default: %s)
+ -s, --slice=SLICE Constrain monitoring to these slices (default: all)
-p, --period=SECONDS Interval in seconds over which to enforce average byte limits (default: %s)
- -h, --help This message
+ -h, --help This message
""".lstrip() % (sys.argv[0], debug, verbose, datafile, format_period(period))
def main():
(version, slices) = pickle.load(f)
f.close()
# Check version of data file
- if version != "$Id: bwmon.py,v 1.3.2.5 2006/08/24 20:16:14 mlhuang Exp $":
+ if version != "$Id: bwmon.py,v 1.15 2006/12/13 21:39:23 faiyaza Exp $":
print "Not using old version '%s' data file %s" % (version, datafile)
raise Exception
except Exception:
- version = "$Id: bwmon.py,v 1.3.2.5 2006/08/24 20:16:14 mlhuang Exp $"
+ version = "$Id: bwmon.py,v 1.15 2006/12/13 21:39:23 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. Global.
nm = NM()
live = []
+ # Get actuall running values from tc.
for params in bwlimit.get():
(xid, share,
minrate, maxrate,
# Monitor only the specified slices
if names and name not in names:
continue
-
- #slices is populated from the pickle file
- #xid is populated from bwlimit (read from /etc/passwd)
+ #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 \
# Faiyaz Ahmed <faiyaza@cs.princeton.edu>
# Copyright (C) 2004-2006 The Trustees of Princeton University
#
-# $Id: swapmon.py,v 1.5.2.4 2006/08/24 20:16:14 mlhuang Exp $
+# $Id: swapmon.py,v 1.11 2006/12/02 19:11:47 mlhuang Exp $
#
import syslog
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.
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
reset_thresh = int(optval)
elif opt == "--reboot-thresh":
reboot_thresh = int(optval)
+ elif opt == "--min-thresh":
+ 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 1.5.2.4 2006/08/24 20:16:14 mlhuang Exp $":
+ if version != "$Id: swapmon.py,v 1.11 2006/12/02 19:11:47 mlhuang Exp $":
print "Not using old version '%s' data file %s" % (version, datafile)
raise Exception
params = {'hostname': socket.gethostname(),
'date': time.asctime(time.gmtime()) + " GMT",
- 'table': summary(total_rss)}
+ 'table': summary(slices, total_mem, total_swap)}
if debug:
print rebooted_subject % params
# Delete data file
os.unlink(datafile)
except Exception:
- version = "$Id: swapmon.py,v 1.5.2.4 2006/08/24 20:16:14 mlhuang Exp $"
+ version = "$Id: swapmon.py,v 1.11 2006/12/02 19:11:47 mlhuang 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
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 slice['rss'] < rss_min:
continue
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:
- 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)
+ # 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)
- slices = slicestat(names)
+ # Reset slice
+ if not resetlist.has_key(slice['name']):
+ resetlist[slice['name']] = Reset(slice['name'])
+ resetlist[slice['name']].reset(params)
+ slices = slicestat(names)
if timer <= 0 or used >= (last_used + change_thresh):
if used >= (last_used + change_thresh):
git://git.onelab.eu / mom.git / commitdiff