X-Git-Url: http://git.onelab.eu/?p=mom.git;a=blobdiff_plain;f=swapmon.py;h=a29a59e12d3f74ad163a695a87e0b30285af20bb;hp=3f6304ec8109eab639db5bd1bf040e7d2d9e69b2;hb=d12d2ba5e6dce633dc95fd53f07448f77ac0fe9c;hpb=8fb15ad132dbaade139b5c41b8d85d6a8fcca016 diff --git a/swapmon.py b/swapmon.py index 3f6304e..a29a59e 100755 --- a/swapmon.py +++ b/swapmon.py @@ -7,9 +7,10 @@ # # Mark Huang # Andy Bavier +# Faiyaz Ahmed # Copyright (C) 2004-2006 The Trustees of Princeton University # -# $Id: swapmon.py,v 1.4 2006/05/02 17:23:14 mlhuang Exp $ +# $Id: swapmon.py,v 1.5 2006/05/09 03:23:57 mlhuang Exp $ # import syslog @@ -44,11 +45,20 @@ period = 30 change_thresh = 5 # Swap utilization at which the largest consumer of physical memory is reset -reset_thresh = 85 +reset_thresh = 75 # Swap utilization at which the machine is rebooted reboot_thresh = 95 +# Time to wait before checking slice again after reset +reset_timeout = 10 + +# Number of strikes before killing (strike, strike, kill) +kill_thresh = 2 + +# Time to wait before removing slice from kill queue +kill_timeout = 30 + # Minimum physical memory utilization to be considered the largest consumer min_thresh = 10 @@ -108,6 +118,117 @@ behavior. %(date)s %(hostname)s alarm %(slice)s """.lstrip() +# Message sent after a slice has been killed +kill_subject = "pl_mom killed slice %(slice)s on %(hostname)s" +kill_body = \ +""" +Sometime before %(date)s, swap space was +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. + +Please reply to this message explaining the nature of your experiment, +and what you are doing to address the problem. + +%(slice)s processes prior to reset: + +%(table)s + +%(date)s %(hostname)s reset %(slice)s +""".lstrip() + + + +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 + """ + + def __init__(self,name): + self.name = name + self.resettimeleft = reset_timeout + self.resetcount = 0 + self.killtimeleft = kill_timeout + + 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 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 + else: + # Once out of probation period (killtimeleft), remove strike + 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. + 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() + 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) + 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 + # 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 + 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 + slicemail(self.name, reset_subject % params, reset_body % params) + + + def usage(): print """ Usage: %s [OPTIONS]... @@ -349,7 +470,7 @@ def main(): (version, slices) = pickle.load(f) f.close() # Check version of data file - if version != "$Id: swapmon.py,v 1.4 2006/05/02 17:23:14 mlhuang Exp $": + if version != "$Id: swapmon.py,v 1.5 2006/05/09 03:23:57 mlhuang Exp $": print "Not using old version '%s' data file %s" % (version, datafile) raise Exception @@ -366,7 +487,7 @@ def main(): # Delete data file os.unlink(datafile) except Exception: - version = "$Id: swapmon.py,v 1.4 2006/05/02 17:23:14 mlhuang Exp $" + version = "$Id: swapmon.py,v 1.5 2006/05/09 03:23:57 mlhuang Exp $" slices = {} # Query process table every 30 seconds, or when a large change in @@ -378,11 +499,18 @@ def main(): # System slices that we have warned but could not reset warned = [] + # Slices that were reset + resetlist = {} + while True: used = swap_used() + + for resetslice in resetlist.keys(): + resetlist[resetslice].update() + if last_used is None: last_used = used - if verbose: + if verbose: print "%d%% swap consumed" % used if used >= reboot_thresh: @@ -404,9 +532,10 @@ def main(): slicelist.sort(lambda a, b: b['rss'] - a['rss']) for slice in slicelist: percent = 100. * slice['rss'] / total_rss + if percent < min_thresh: continue - + print "%d%% swap consumed, slice %s is using %s (%d%%) of memory" % \ (used, slice['name'], @@ -444,25 +573,10 @@ def main(): print "Warning slice " + slice['name'] slicemail(slice['name'], alarm_subject % params, alarm_body % params) else: - # Otherwise, reset - if debug: - print reset_subject % params - print reset_body % params - else: - try: - pid = os.fork() - if pid == 0: - print "Resetting slice " + slice['name'] - vserver = VServer(slice['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:" % slice['name'], err - slicemail(slice['name'], reset_subject % params, reset_body % params) - break + # Reset slice + if not resetlist.has_key(slice['name']): + resetlist[slice['name']] = Reset(slice['name']) + resetlist[slice['name']].reset(params) elif timer <= 0 or used >= (last_used + change_thresh): if used >= (last_used + change_thresh):