X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=swapmon.py;h=da8f3a0129941471731ea3969e02e81ed371383d;hb=80c18c64693ca278b9731bd2c44b925fcb298d00;hp=37743b17a6f60f3cad11d178f4f5c09896c6a5d2;hpb=662dbcbe3e431c6dc9755757428fb7f837a21498;p=mom.git diff --git a/swapmon.py b/swapmon.py index 37743b1..da8f3a0 100755 --- a/swapmon.py +++ b/swapmon.py @@ -10,7 +10,7 @@ # Faiyaz Ahmed # Copyright (C) 2004-2006 The Trustees of Princeton University # -# $Id: swapmon.py,v 1.10 2006/08/16 16:18:45 faiyaza Exp $ +# $Id$ # import syslog @@ -51,7 +51,7 @@ reset_thresh = 80 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 @@ -97,6 +97,8 @@ 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. +http://summer.cs.princeton.edu/status/tabulator.cgi?table=slices/table_%(slice)s + %(slice)s processes prior to reset: %(table)s @@ -146,107 +148,104 @@ and what you are doing to address the problem. 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.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 - - - # 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: - 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 (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. - 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() + """ + 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.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 + + + # 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: + 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 (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. + 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 """ @@ -272,13 +271,13 @@ 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 @@ -362,7 +361,7 @@ def slicestat(names = None): slice['procs'].append(proc) slice['vsize'] += proc['vsize'] - slice['sz'] += proc['sz'] + slice['sz'] += proc['sz'] slice['rss'] += proc['rss'] slices[proc['xid']] = slice @@ -373,32 +372,27 @@ def memtotal(): """ 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: - (name, value, kb) = line.split() - except: - continue - if name == "MemTotal:": - mem = int(value) - elif name == "SwapTotal:": - swap = int(value) + try: + (name, value, kb) = line.split() + except: + continue + if name == "MemTotal:": + mem = int(value) + 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 @@ -412,21 +406,21 @@ def swap_used(): 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: - (total_mem, total_swap) = memtotal() + (total_mem, total_swap) = memtotal() table = "%-20s%10s%24s%24s\n\n" % ("Slice", "Processes", "Memory Usage", "Potential Usage") for slice in slicelist: @@ -436,10 +430,79 @@ def summary(slices = None, total_mem = None, total_swap = 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 + (version, slices) = pickle.load(f) + f.close() + # Check version of data file + if version != "$Id$": + 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 = "$Id$" + 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 @@ -497,33 +560,7 @@ def main(): # 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 1.10 2006/08/16 16:18:45 faiyaza Exp $": - print "Not using old version '%s' data file %s" % (version, datafile) - raise Exception - - params = {'hostname': socket.gethostname(), - 'date': time.asctime(time.gmtime()) + " GMT", - 'table': summary(slices, total_mem, total_swap)} - - if debug: - print rebooted_subject % params - print rebooted_body % params - else: - slicemail(None, rebooted_subject % params, rebooted_body % params) - - # Delete data file - os.unlink(datafile) - except Exception: - version = "$Id: swapmon.py,v 1.10 2006/08/16 16:18:45 faiyaza Exp $" - slices = {} + slices = readdat() # Query process table every 30 seconds, or when a large change in # swap utilization is detected. @@ -539,98 +576,72 @@ def main(): while True: used = swap_used() - - for resetslice in resetlist.keys(): - resetlist[resetslice].update() - - if last_used is None: - last_used = used - - if verbose: - print "%d%% swap consumed" % used - + 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 - if verbose: - print "Saving %s" % datafile - f = open(datafile, "w") - pickle.dump((version, slices), f) - f.close() - + writedat(slices) # Goodbye, cruel world print "%d%% swap consumed, rebooting" % used - if not debug: - bwlimit.run("/bin/sync; /sbin/reboot -f") - + 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() + # 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 - + 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) - + (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} - + # 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]) - + is_system_slice = filter(None, + [re.match(pattern, slice['name']) for pattern in system_slices]) + + # Do not reset system slices, just warn once 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 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: - # 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)