Changeset 2416 for palm/trunk/SCRIPTS

Timestamp:

Sep 6, 2017 2:28:14 PM (7 years ago)

Author:

maronga

Message:

watchdog adapted for palmrun

Location:

palm/trunk/SCRIPTS

Files:

• palm_wd (modified) (9 diffs)
• palm_wd_files/.wdd.config.hlrnIII (modified) (1 diff)
• palm_wdd (modified) (10 diffs)

palm/trunk/SCRIPTS/palm_wd

@@ -25 +25 @@
 # -----------------
 # $Id$
+# Adapted for palmrun
+# 
+# 1754 2016-02-22 13:50:22Z maronga
 #
 # 1753 2016-02-22 13:49:49Z maronga
@@ -662 +665 @@
         # Return internal jobname
         jobname = self.table.item(self.table.currentRow(),6).text()
+        jobrealname = self.table.item(self.table.currentRow(),0).text()     
  
         # Check description of job in order to login on the correct host
@@ -700 +704 @@
     def ShowDetails(self):
 
-        jobname = self.table.item(self.table.currentRow(),6).text()
+        jobname = self.table.item(self.table.currentRow(),6).text()  
         descr   = self.table.item(self.table.currentRow(),1).text()
         for h in range(0,len(description)):
@@ -725 +729 @@
         # Return internal jobname
         jobname = self.table.item(self.table.currentRow(),6).text()
- 
+        jobrealname = self.table.item(self.table.currentRow(),0).text()   
         # Check description of job in order to login on the correct host
         descr = self.table.item(self.table.currentRow(),1).text()
@@ -733 +737 @@
                 user = username[h]
 
-        ssh = sub.Popen(["ssh", "%s" % host, "/sw/tools/python/2.7.6/generic/bin/python palm_wdd stop " + user + " " + jobname],
+        ssh = sub.Popen(["ssh", "%s" % host, "/sw/tools/python/2.7.6/generic/bin/python palm_wdd stop " + jobrealname],
                            shell=False,
                            stdout=sub.PIPE,
@@ -756 +760 @@
         # Return internal jobname
         jobname = self.table.item(self.table.currentRow(),6).text()
- 
+        jobrealname = self.table.item(self.table.currentRow(),0).text()   
+         
         # Check description of job in order to login on the correct host
         descr = self.table.item(self.table.currentRow(),1).text()
@@ -764 +769 @@
                 user = username[h]
 
-        ssh = sub.Popen(["ssh", "%s" % host, "/sw/tools/python/2.7.6/generic/bin/python palm_wdd restart " + user + " " + jobname],
+        ssh = sub.Popen(["ssh", "%s" % host, "/sw/tools/python/2.7.6/generic/bin/python palm_wdd restart " + jobrealname],
                            shell=False,
                            stdout=sub.PIPE,
@@ -796 +801 @@
                 user = username[h]
 
-        ssh = sub.Popen(["ssh", "%s" % host, "/sw/tools/python/2.7.6/generic/bin/python palm_wdd parin " + user + " " + jobname],
+        ssh = sub.Popen(["ssh", "%s" % host, "/sw/tools/python/2.7.6/generic/bin/python palm_wdd parin " + jobrealname],
                            shell=False,
                            stdout=sub.PIPE,
@@ -829 +834 @@
                 user = username[h]
 
-        ssh = sub.Popen(["ssh", "%s" % host, "/sw/tools/python/2.7.6/generic/bin/python palm_wdd rc " + user + " " + jobname],
+        ssh = sub.Popen(["ssh", "%s" % host, "/sw/tools/python/2.7.6/generic/bin/python palm_wdd rc " + jobrealname],
                            shell=False,
                            stdout=sub.PIPE,

palm/trunk/SCRIPTS/palm_wd_files/.wdd.config.hlrnIII

@@ -5 +5 @@
 [Settings]
 readqueue="showq | egrep"
-tmpdir="/gfs1/tmp/"
+tmpdir="/gfs1/work/"
 canceljob="canceljob"
 checkjob="checkjob"

palm/trunk/SCRIPTS/palm_wdd

@@ -20 +20 @@
 # Current revisions:
 # -----------------
-#
+# 
 # 
 # Former revisions:
 # -----------------
 # $Id$
-#
-# Bugfix: return proper error messages when .wdd.config is not found and for
-# wrong actions
+# Adapted for palmrun
+# 
+# 1619 2015-07-13 06:53:19Z maronga
 #
 # 1618 2015-07-13 06:52:15Z maronga
@@ -65 +65 @@
 import ConfigParser
 import os
+import pwd
 from subprocess import check_output
 import sys
-import socket
+
 
 # Read configuration file
 # First check if the configuration file exists
 if ( os.path.exists('.wdd.config') == False ):
-    sys.stderr.write("No configuration file .wdd.config found on remote host %s." % socket.gethostname())
+    print "Error. No configuration file .wdd.config found."
     raise SystemExit     
 
@@ -89 +90 @@
 action   = str(sys.argv[1])
 data     = str(sys.argv[2])
-if ( len(sys.argv) > 3 ):
-   data2 =  str(sys.argv[3])
-
-
-cmd_readqueue = cmd_readqueue + " " + data
-cmd_tmpdir    = cmd_tmpdir + data
+
+cmd_readqueue = cmd_readqueue + " " + pwd.getpwuid( os.getuid() )[ 0 ]
+cmd_tmpdir    = cmd_tmpdir + pwd.getpwuid( os.getuid() )[ 0 ]
 
 # reading queuing system
@@ -137 +135 @@
             out = check_output(cmd_progress, shell=True, stderr=devnull)
             progress_lines = out.splitlines()
-            job_progress = progress_lines[1].split(" ")[1]
+            job_progress = progress_lines[1].split(" ")[3]
             out = None
          except:
@@ -192 +190 @@
 
 
-def DoStopNow(username,jobid):
-
-   # collect progress information
-   cmd_dostop = "touch " + cmd_tmpdir + "/" + username + "." + jobid.partition(".")[2] + "/DO_STOP_NOW"
+def DoStopNow(jobid):
+
+   # collect progress information
+   cmd_dostop = "touch " + cmd_tmpdir + "/" + jobid + "/DO_STOP_NOW"
    try: 
       devnull = open(os.devnull, 'w')
@@ -204 +202 @@
       return return_message
 
-def DoRestartNow(username,jobid):
-
-   # collect progress information
-   cmd_dorestart = "touch " + cmd_tmpdir + "/" + username + "." + jobid.partition(".")[2] + "/DO_RESTART_NOW"
+def DoRestartNow(jobid):
+
+   # collect progress information
+   cmd_dorestart = "touch " + cmd_tmpdir + "/" + jobid + "/DO_RESTART_NOW"
    try: 
       devnull = open(os.devnull, 'w')
@@ -216 +214 @@
       return return_message
 
-def GetPARIN(username,jobid):
-
-   # collect progress information
-   cmd_dorestart = "cat " + cmd_tmpdir + "/" + username + "." + jobid.partition(".")[2] + "/PARIN"
+def GetPARIN(jobid):
+
+   # collect progress information
+   cmd_dorestart = "cat " + cmd_tmpdir + "/" + jobid + "/PARIN"
    try: 
       devnull = open(os.devnull, 'w')
@@ -226 +224 @@
       out = None
    except:
-      return_message = "Action failed."  
+      return_message = "Action failed."   + "cat " + cmd_tmpdir + "/" + jobid + "/PARIN" 
       
    return return_message
 
-def GetRC(username,jobid):
-
-   # collect progress information
-   cmd_dorestart = "cat " + cmd_tmpdir + "/" + username + "." + jobid.partition(".")[2] + "/RUN_CONTROL"
+def GetRC(jobid):
+
+   # collect progress information
+   cmd_dorestart = "cat " + cmd_tmpdir + "/" + jobid + "/RUN_CONTROL"
    try: 
       devnull = open(os.devnull, 'w')
@@ -240 +238 @@
       out = None
    except:
-      return_message = "Action failed."  
+      return_message = "Action failed."
       
    return return_message
@@ -254 +252 @@
    print GetStartTime(data)
 elif ( action == "stop"):
-   print DoStopNow(data,data2)
+   print DoStopNow(data)
 elif ( action == "restart"):
-   print DoRestartNow(data,data2)
+   print DoRestartNow(data)
 elif ( action == "parin"):
-   print GetPARIN(data,data2)
+   print GetPARIN(data)
 elif ( action == "rc"):
-   print GetRC(data,data2)  
+   print GetRC(data)  
 else:
-   sys.stderr.write("Unknown action on remote host %s." % hostname)
-
+   print "Error. Action " + action + " unknown."
+