3 # Swap monitoring daemon. Every 30 seconds, checks process memory
4 # usage. At 90% utilization, resets the slice that is consuming the
5 # most physical memory. At 95% utilization, reboots the machine to
8 # Mark Huang <mlhuang@cs.princeton.edu>
9 # Andy Bavier <acb@cs.princeton.edu>
10 # Copyright (C) 2004-2006 The Trustees of Princeton University
12 # $Id: swapmon.py,v 1.1 2006/04/28 19:26:59 mlhuang Exp $
24 # util-vserver/python/vserver.py allows us to control slices directly
26 from vserver import VServer
28 # bwlimit exports a few useful functions like run(), get_xid(), and get_slice()
37 datafile = "/var/lib/misc/swapmon.dat"
39 # Seconds between process analysis
42 # Minimum change in swap utilization over 30 seconds that will trigger
43 # early process analysis.
46 # Swap utilization at which the largest consumer of physical memory is reset
49 # Swap utilization at which the machine is rebooted
52 # Minimum physical memory utilization to be considered the largest consumer
55 # System slices that should not be reset (regexps)
56 system_slices = ['root', PLC_SLICE_PREFIX + '_']
58 # Message sent after a critical reboot
59 rebooted_subject = "pl_mom rebooted %(hostname)s"
62 Sometime before %(date)s, swap space was
63 nearly exhausted on %(hostname)s, so pl_mom rebooted it.
65 Slices active prior to reboot are listed below. Memory usage
66 statistics are not entirely accurate due to threading.
70 %(date)s %(hostname)s reboot
73 # Message sent after a hog is reset
74 reset_subject = "pl_mom reset slice %(slice)s on %(hostname)s"
77 Sometime before %(date)s, swap space was
78 nearly exhausted on %(hostname)s.
80 Slice %(slice)s was reset since it was the largest consumer of
81 physical memory at %(rss)s (%(percent)4.1f%%).
83 Please reply to this message explaining the nature of your experiment,
84 and what you are doing to address the problem.
86 %(slice)s processes prior to reset:
90 %(date)s %(hostname)s reset %(slice)s
93 # Message sent to system slices that should not be reset
94 alarm_subject = "pl_mom alarm slice %(slice)s on %(hostname)s"
97 Sometime before %(date)s, swap space was
98 nearly exhausted on %(hostname)s.
100 System slice %(slice)s was the largest consumer of physical memory at
101 %(rss)s (%(percent)4.1f%%). It was not reset, but please verify its
104 %(slice)s processes prior to alarm:
108 %(date)s %(hostname)s alarm %(slice)s
113 Usage: %s [OPTIONS]...
116 -d, --debug Enable debugging (default: %s)
117 -v, --verbose Increase verbosity level (default: %d)
118 -f, --file=FILE Data file (default: %s)
119 -s, --slice=SLICE Constrain monitoring to these slices (default: all)
120 -p, --period=SECONDS Seconds between normal process analysis (default: %s)
121 --reset-thresh=PERCENT Swap utilization at which slice reset is attempted
122 --reboot-thresh=PERCENT Swap utilization at which the machine is rebooted
123 --min-thresh=PERCENT Minimum physical memory utilization to be considered a hog
124 --system-slice=SLICE System slice that should not be reset
125 -h, --help This message
126 """.lstrip() % (sys.argv[0], debug, verbose, datafile, format_period(period))
128 def slicestat(names = None):
130 Get status of specified slices (if names is None or empty, all
131 slices). vsize and rss are in KiB. Returns
133 {xid: {'xid': slice_id,
135 'procs': [{'pid': pid, 'xid': slice_id, 'user', username, 'cmd': command,
136 'vsize': virtual_kib, 'rss': physical_kib,
137 'pcpu': cpu_percent, 'pmem': mem_percent}]
138 'vsize': total_virtual_kib,
139 'rss': total_physical_kib}}
142 # Mandatory fields. xid is a virtual field inserted by vps. Make
143 # sure cmd is last so that it does not get truncated
145 fields = ['pid', 'xid', 'user', 'vsize', 'rss', 'pcpu', 'pmem', 'cmd']
147 # vps inserts xid after pid in the output, but ps doesn't know
148 # what the field means.
149 ps_fields = list(fields)
150 ps_fields.remove('xid')
154 # Eat the header line. vps depends on the header to figure out
155 # which column is the PID column, so we can't just tell ps not to
157 for line in bwlimit.run("/usr/sbin/vps -e -o " + ",".join(ps_fields))[1:]:
161 # Replace "0 MAIN" and "1 ALL_PROC" (the special monikers that
162 # vps uses to denote the root context and the "all contexts"
163 # context) with "0" so that we can just split() on whitespace.
164 line = line.replace("0 MAIN", "0").replace("1 ALL_PROC", "0")
166 # Represent process as a dict of fields
167 values = line.split(None, len(fields) - 1)
168 if len(values) != len(fields):
170 proc = dict(zip(fields, values))
172 # Convert ints and floats
175 proc[field] = int(proc[field])
178 proc[field] = float(proc[field])
182 # Assign (pl_)sshd processes to slice instead of root
183 m = re.search(r"sshd: ([a-zA-Z_]+)", proc['cmd'])
185 xid = bwlimit.get_xid(m.group(1))
189 name = bwlimit.get_slice(proc['xid'])
191 # Orphaned (not associated with a slice) class
192 name = "%d?" % proc['xid']
194 # Monitor only the specified slices
195 if names and name not in names:
198 # Additional overhead calculations from slicestat
200 # Include 12 KiB of process overhead =
201 # 4 KiB top-level page table +
202 # 4 KiB kernel structure +
203 # 4 KiB basic page table
206 # Include additional page table overhead
207 if proc['vsize'] > 4096:
208 proc['rss'] += 4 * ((proc['vsize'] - 1) / 4096)
210 if slices.has_key(proc['xid']):
211 slice = slices[proc['xid']]
213 slice = {'xid': proc['xid'], 'name': name, 'procs': [], 'vsize': 0, 'rss': 0}
215 slice['procs'].append(proc)
216 slice['vsize'] += proc['vsize']
217 slice['rss'] += proc['rss']
219 slices[proc['xid']] = slice
225 Returns total physical memory on the system in KiB.
228 meminfo = open("/proc/meminfo", "r")
229 line = meminfo.readline()
231 if line[0:8] == "MemTotal":
232 # MemTotal: 255396 kB
233 (name, value, kb) = line.split()
240 Returns swap utilization on the system as a whole percentage (0-100).
247 swaps = open("/proc/swaps", "r")
249 lines = swaps.readlines()[1:]
252 # /dev/mapper/planetlab-swap partition 1048568 3740 -1
253 (filename, type, size, used, priority) = line.strip().split()
255 total_swap += int(size)
256 total_used += int(used)
257 except ValueEror, err:
259 except (IOError, KeyError), err:
262 return 100 * total_used / total_swap
266 global debug, verbose, datafile
267 global period, change_thresh, reset_thresh, reboot_thresh, min_thresh, system_slices
272 longopts = ["debug", "verbose", "file=", "slice=", "help"]
273 longopts += ["period=", "reset-thresh=", "reboot-thresh=", "min-thresh=", "system-slice="]
274 (opts, argv) = getopt.getopt(sys.argv[1:], "dvf:s:ph", longopts)
275 except getopt.GetoptError, err:
276 print "Error: " + err.msg
280 for (opt, optval) in opts:
281 if opt == "-d" or opt == "--debug":
283 elif opt == "-v" or opt == "--verbose":
285 elif opt == "-f" or opt == "--file":
287 elif opt == "-s" or opt == "--slice":
289 elif opt == "-p" or opt == "--period":
291 elif opt == "--change-thresh":
292 change_thresh = int(optval)
293 elif opt == "--reset-thresh":
294 reset_thresh = int(optval)
295 elif opt == "--reboot-thresh":
296 reboot_thresh = int(optval)
297 elif opt == "--min-thresh":
298 min_thresh = int(optval)
299 elif opt == "--system-slice":
300 system_slices.append(optval)
305 # Check if we are already running
312 # Redirect stdout and stderr to syslog
313 syslog.openlog("swapmon")
314 sys.stdout = sys.stderr = Logger()
316 # Get total physical memory
317 total_rss = memtotal()
320 f = open(datafile, "r+")
322 print "Loading %s" % datafile
323 (version, slices) = pickle.load(f)
325 # Check version of data file
326 if version != "$Id: swapmon.py,v 1.1 2006/04/28 19:26:59 mlhuang Exp $":
327 print "Not using old version '%s' data file %s" % (version, datafile)
330 # Send notification if we rebooted the node because of swap exhaustion
331 slicelist = slices.values()
332 slicelist.sort(lambda a, b: b['rss'] - a['rss'])
334 table = "%-20s%10s%24s\n\n" % ("Slice", "Processes", "Memory Usage")
335 for slice in slicelist:
336 table += "%-20s%10d%16s (%4.1f%%)\n" % \
337 (slice['name'], len(slice['procs']),
338 format_bytes(slice['rss'] * 1024, si = False),
339 100. * slice['rss'] / total_rss)
341 params = {'hostname': socket.gethostname(),
342 'date': time.asctime(time.gmtime()) + " GMT",
346 print rebooted_subject % params
347 print rebooted_body % params
349 slicemail(None, rebooted_subject % params, rebooted_body % params)
354 version = "$Id: swapmon.py,v 1.1 2006/04/28 19:26:59 mlhuang Exp $"
357 # Query process table every 30 seconds, or when a large change in
358 # swap utilization is detected.
363 # System slices that we have warned but could not reset
368 if last_used is None:
371 print "%d%% swap consumed" % used
373 if used >= reboot_thresh:
374 # Dump slice state before rebooting
376 print "Saving %s" % datafile
377 f = open(datafile, "w")
378 pickle.dump((version, slices), f)
381 # Goodbye, cruel world
382 print "%d%% swap consumed, rebooting" % used
384 bwlimit.run("/bin/sync; /sbin/reboot -f")
386 elif used >= reset_thresh:
388 slicelist = slices.values()
389 slicelist.sort(lambda a, b: b['rss'] - a['rss'])
390 for slice in slicelist:
391 percent = 100. * slice['rss'] / total_rss
392 if percent < min_thresh:
395 print "%d%% swap consumed, slice %s is using %s (%d%%) of memory" % \
398 format_bytes(slice['rss'] * 1024, si = False),
401 slice['procs'].sort(lambda a, b: b['rss'] - a['rss'])
403 table = "%5s %10s %10s %4s %4s %s\n\n" % ("PID", "VIRT", "RES", '%CPU', '%MEM', 'COMMAND')
404 for proc in slice['procs']:
405 table += "%5s %10s %10s %4.1f %4.1f %s\n" % \
407 format_bytes(proc['vsize'] * 1024, si = False),
408 format_bytes(proc['rss'] * 1024, si = False),
409 proc['pcpu'], proc['pmem'], proc['cmd'])
411 params = {'hostname': socket.gethostname(),
412 'date': time.asctime(time.gmtime()) + " GMT",
414 'slice': slice['name'],
415 'rss': format_bytes(slice['rss'] * 1024, si = False),
418 # Match slice name against system slice patterns
419 is_system_slice = filter(None, [re.match(pattern, slice['name']) for pattern in system_slices])
422 # Do not reset system slices, just warn once
423 if slice['name'] not in warned:
424 warned.append(slice['name'])
426 print alarm_subject % params
427 print alarm_body % params
429 print "Warning slice " + slice['name']
430 slicemail(slice['name'], alarm_subject % params, alarm_body % params)
434 print reset_subject % params
435 print reset_body % params
440 print "Resetting slice " + slice['name']
441 vserver = VServer(slice['name'])
443 vserver.start(wait = False)
447 except Exception, err:
448 print "Warning: Exception received while resetting slice %s:" % slice['name'], err
449 slicemail(slice['name'], reset_subject % params, reset_body % params)
452 elif timer <= 0 or used >= (last_used + change_thresh):
453 if used >= (last_used + change_thresh):
454 print "%d%% swap consumed, %d%% in last %d seconds" % \
455 (used, used - last_used, period - timer)
457 slices = slicestat(names)
460 # Keep track of large changes in swap utilization
468 if __name__ == '__main__':