X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=swapmon.py;h=3f6304ec8109eab639db5bd1bf040e7d2d9e69b2;hb=36d48e6b65c5b4514e417f425aee54685989a605;hp=7d579e4766fbd119724d161e14648ad9916f0b81;hpb=42dbe689a93025034f3e798174e54f3e2a0d3371;p=mom.git

diff --git a/swapmon.py b/swapmon.py
index 7d579e4..3f6304e 100755
--- a/swapmon.py
+++ b/swapmon.py
@@ -9,7 +9,7 @@
 # Andy Bavier <acb@cs.princeton.edu>
 # Copyright (C) 2004-2006 The Trustees of Princeton University
 #
-# $Id: BandwidthMonitor.py,v 1.1 2006/04/25 14:40:28 mlhuang Exp $
+# $Id: swapmon.py,v 1.4 2006/05/02 17:23:14 mlhuang Exp $
 #
 
 import syslog
@@ -21,9 +21,6 @@ import pickle
 import socket
 import time
 
-import textwrap
-wrap = textwrap.TextWrapper()
-
 # util-vserver/python/vserver.py allows us to control slices directly
 # from Python
 from vserver import VServer
@@ -125,6 +122,7 @@ Options:
         --reboot-thresh=PERCENT Swap utilization at which the machine is rebooted
         --min-thresh=PERCENT    Minimum physical memory utilization to be considered a hog
         --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))
 
@@ -182,6 +180,12 @@ def slicestat(names = None):
                 except ValueError:
                     pass
 
+        # vps sometimes prints ERR instead of a context ID if it
+        # cannot identify the context of an orphaned (usually dying)
+        # process. Skip these processes.
+        if type(proc['xid']) != int:
+            continue
+
         # Assign (pl_)sshd processes to slice instead of root
         m = re.search(r"sshd: ([a-zA-Z_]+)", proc['cmd'])
         if m is not None:
@@ -264,6 +268,22 @@ def swap_used():
 
     return 100 * total_used / total_swap
 
+def summary(names = None, total_rss = memtotal()):
+    """
+    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")
+    for slice in slicelist:
+        table += "%-20s%10d%16s (%4.1f%%)\n" % \
+                 (slice['name'], len(slice['procs']),
+                  format_bytes(slice['rss'] * 1024, si = False),
+                  100. * slice['rss'] / total_rss)
+
+    return table
+
 def main():
     # Defaults
     global debug, verbose, datafile
@@ -272,7 +292,7 @@ def main():
     names = []
 
     try:
-        longopts = ["debug", "verbose", "file=", "slice=", "help"]
+        longopts = ["debug", "verbose", "file=", "slice=", "status", "help"]
         longopts += ["period=", "reset-thresh=", "reboot-thresh=", "min-thresh=", "system-slice="]
         (opts, argv) = getopt.getopt(sys.argv[1:], "dvf:s:ph", longopts)
     except getopt.GetoptError, err:
@@ -301,6 +321,9 @@ def main():
             min_thresh = int(optval)
         elif opt == "--system-slice":
             system_slices.append(optval)
+        elif opt == "--status":
+            print summary(names)
+            sys.exit(0)
         else:
             usage()
             sys.exit(0)
@@ -326,24 +349,13 @@ def main():
         (version, slices) = pickle.load(f)
         f.close()
         # Check version of data file
-        if version != "$Id: bwmon.py,v 1.1 2006/04/25 14:40:28 mlhuang Exp $":
+        if version != "$Id: swapmon.py,v 1.4 2006/05/02 17:23:14 mlhuang Exp $":
             print "Not using old version '%s' data file %s" % (version, datafile)
             raise Exception
 
-        # Send notification if we rebooted the node because of swap exhaustion
-        slicelist = slices.values()
-        slicelist.sort(lambda a, b: b['rss'] - a['rss'])
-
-        table = "%-20s%10s%24s\n\n" % ("Slice", "Processes", "Memory Usage")
-        for slice in slicelist:
-            table += "%-20s%10d%16s (%4.1f%%)\n" % \
-                     (slice['name'], len(slice['procs']),
-                      format_bytes(slice['rss'] * 1024, si = False),
-                      100. * slice['rss'] / total_rss)
-
         params = {'hostname': socket.gethostname(),
                   'date': time.asctime(time.gmtime()) + " GMT",
-                  'table': table}            
+                  'table': summary(total_rss)}
 
         if debug:
             print rebooted_subject % params
@@ -354,7 +366,7 @@ def main():
         # Delete data file
         os.unlink(datafile)
     except Exception:
-        version = "$Id: bwmon.py,v 1.1 2006/04/25 14:40:28 mlhuang Exp $"
+        version = "$Id: swapmon.py,v 1.4 2006/05/02 17:23:14 mlhuang Exp $"
         slices = {}
 
     # Query process table every 30 seconds, or when a large change in